ProgrammableMarketing/xdof.ai (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 6
  • This quarter · 9
  • Big bets · 3
  • Wildcards · 3
  • How we can help
Generated Oct 23, 2026
  1. Programmable
  2. Playbooks
  3. XDOF
XDOF logo

growth playbook · Oct 23, 2026

xdof.ai (opens in a new tab)

XDOF growth playbook

Diagnosis

XDOF has rare assets: co-founders who authored GELLO, the most downloaded open teleoperation dataset (ABC-130K, 1.9 million downloads), a robot fleet that has produced 100,000s of hours of data, and named backers.

Its hiring shows the next phase: a founding Marketing Lead, a product for self-serve evaluations, data venues in Latin America and a marketing and operations team in Indonesia. The website has not caught up.

The homepage's only button is for jobs and its text loads by script, there is no page for the core data service, the blog went quiet in June and the dataset audience is never invited to buy. This playbook connects the dataset audience to Evaluations first, rebuilds the homepage for three audiences, then adds a leaderboard, self-serve evals, local partner pages and a data-quality report.

Growth Score

6/10

Biggest Opportunity

ABC-130K's 234,346 monthly downloads come from the researchers and labs who buy evaluations and data; inviting them in is the fastest path to pipeline.

Biggest Risk

Larger data vendors such as Scale AI and Encord are marketing hard to physical AI teams, and XDOF's site does not yet explain its data service.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    Turn ABC-130K's Hugging Face audience into evaluation and data customers

    Add the "From the team" section and tracked link to the dataset card today.

    High impact First result in 2 weeks
  2. 2

    Give the homepage a front door for each of the three audiences: labs, data partners and hires

    Rewrite the homepage title and description today.

    High impact First result in 2 weeks
  3. 3

    Publish a page for the data collection service, the core business with no page

    Draft the page outline with the operations team.

    High impact First result in 1 month

02

Your idea portfolio

21 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 8 ideas are marked Unexpected: plays you are unlikely to have tried.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 1Big bets · 9Fill-ins · 7Reconsider · 4
Effort →

Showing 21 of 21 ideas

This week 6

Quick wins with a clear first action.

  • 01

    ABC-130K was downloaded 234,346 times on Hugging Face in the last month (1.9 million in total), and the gate shares each requester's username and email with XDOF. These are robotics researchers and lab teams, the exact buyers of XDOF Evaluations and data. Yet the dataset card points only to the paper and code, never to xdof.ai, Evaluations or the "we'd love to talk" offer in the launch post. The founding Marketing Lead is hired to run launches and turn attention into pipeline; this is the largest audience XDOF already has.

    High impactLow effort First result in 2 weeks Marketing Lead

    First action

    Add the "From the team" section and tracked link to the dataset card today.

    How to measure it

    Evaluation inquiries from the dataset

    Now
    –
    Target
    15 a month
    By
    90 days

    Steps

    1. 1

      Edit the card

      Short section with Evaluations and data collection, tracked link.

      Marketing Lead 1 day

    2. 2

      Build the landing page

      /abc for dataset users: evals on ABC tasks, custom data, contact.

      Marketing Lead 3 days

    3. 3

      Add opt-in to the gate

      One checkbox for updates; no change to access.

      Member of Technical Staff, Research Engineer 1 day

    4. 4

      Group requesters

      By organization domain; flag labs with several requesters.

      Marketing Lead 2 days

    5. 5

      Founder notes

      One personal note per lab, offering a first eval round.

      Co-founders 1 week

    6. 6

      Answer the discussions

      Reply to open threads with links where useful.

      Member of Technical Staff, Research Engineer 2 days

    7. 7

      Track

      Landing visits, inquiries, first evals.

      Marketing Lead weekly

    8. 8

      Repeat for each release

      Same section on ABC-Sim and ABC-Eval.

      Marketing Lead per release

    First result
    2 weeks
    Cost
    Staff time only
    Owner
    Marketing Lead

    Smallest test

    Add the card section and count tracked clicks for two weeks.

    Why it works

    Add a short "From the team that collected this data" section to the card: 24-hour evaluations on the same YAM stations, custom data collection, and a tracked link to a dedicated landing page. Add an opt-in question to the gate ("Tell me about new XDOF datasets and evaluations"). Group past requesters by organization domain, and for labs and companies with several requesters, send one personal note from a founder offering a free first evaluation round on ABC tasks.

    The people downloading the dataset have already chosen XDOF's data and hardware setup; an evaluation on the same tasks is the natural next step, and an offer on the card reaches them at the moment of use.

    Ready to use

    ChecklistDataset-to-customer launch list

    [ ] "From the team" section on the ABC-130K card
    [ ] Tracked link to /abc landing page
    [ ] Opt-in checkbox in the access gate
    [ ] Requesters grouped by organization
    [ ] Founder note to labs with several requesters
    [ ] Open discussion threads answered
    [ ] Same section ready for ABC-Sim and ABC-Eval

    Risks and how to handle them

    • Researchers dislike sales email

      Write only to labs with several requesters, from a founder, with a free offer and no sequence.

    • Consent for past requesters

      Check the gate terms first; use the opt-in for everyone from now on.

    Evidence

    • Hugging FaceXDOF/ABC-130k on Hugging Face (October 22, 2026): 234,346 downloads in the last month and 1,924,086 in total, 135 likes, 65 followers of the XDOF organization and 10 discussion threads. Access is gated with automatic approval and shares each requester's username and email with XDOF. The dataset card points to the paper site and code but not to xdof.ai or XDOF Evaluations.Source
    • Your websiteABC-130K post: "over 130,000 episodes across 195 bimanual manipulation tasks" under Apache 2.0; ABC-Sim and ABC-Eval marked "(coming soon)"; episodes viewable on Foundry; built with the UC Berkeley authors of "Scalable Behavior Cloning with Open Data, Training, and Evaluation"; "if you ever need more data or evaluations like this, we'd love to talk".Source
    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
  • 02

    The Marketing Lead must build "a distinct message" for customers, data partners and prospective hires. Today the homepage's only button is "See open roles", its hero text loads by script, and the HTML that crawlers and link previews read contains just "XDOF" and the menu. A lab buyer arriving from a paper has no path to Evaluations or data.

    High impactMedium effort First result in 2 weeks Marketing Lead

    First action

    Rewrite the homepage title and description today.

    How to measure it

    Homepage-to-product click rate

    Now
    –
    Target
    25%
    By
    60 days

    Steps

    1. 1

      Write the line

      What XDOF does in one sentence.

      Marketing Lead 1 day

    2. 2

      Pick proof

      Hours of data, ABC-130K, GELLO, backers.

      Marketing Lead 1 day

    3. 3

      Server-render the hero

      Text in the HTML; animation stays on top.

      Product Engineer 3 days

    4. 4

      Add four paths

      Evaluate, data, partner, join.

      Product Engineer 2 days

    5. 5

      New title and description

      Say what XDOF sells.

      Marketing Lead 1 hour

    6. 6

      Measure

      Clicks per path.

      Marketing Lead weekly

    First result
    2 weeks
    Cost
    Staff time only
    Owner
    Marketing Lead

    Smallest test

    Add the four buttons under the current hero this week.

    Why it works

    Rebuild the homepage hero as server-sent text: one line on what XDOF does (data, evaluations, robot systems for physical AI), proof (100,000s of hours of robot data, ABC-130K, GELLO), and three clear paths: "Evaluate a policy", "Get data" and "Partner with us", with "Join the team" as the fourth. Rewrite the title and description to say what XDOF sells.

    Visitors find their path in seconds, and search and AI assistants can finally read what XDOF does.

    Risks and how to handle them

    • Loses the minimal look

      Keep the animation; add one short text block beneath.

    Evidence

    • Your websiteHomepage (October 22, 2026): title "XDOF" and description "Defining motion for autonomous systems."; the HTML sent to browsers and crawlers holds only the menu and footer. The hero loads by script; its code holds one sentence ("we are a group of engineers and roboticists intensely focused on assembling the ingredients for the first generation of robotics foundation models"), "Backed by Thrive Capital, WndrCo, Andreessen Horowitz (a16z), and Samsung Next." and a "Join the team" panel whose button is "See open roles".Source
    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
    • Your websiteAbout page: XDOF partners "with the world's leading robotics labs and most promising robotics companies to deliver production-scale datasets, advanced robotic systems, and the tooling"; "Our co-founders authored GELLO, now commonly deployed for teleoperation." The launch post adds a team from "Covariant, Tesla, Google, Uber, Applied Intuition" with "roots in Berkeley AI research".Source
  • 03

    Job posts call training data "the bottleneck" and describe XDOF's collection systems, operations and an "exabyte-scale data warehouse"; the About page promises "production-scale datasets". But the only product page is Evaluations, and the sitemap has no page for custom data collection.

    High impactMedium effort First result in 1 month Marketing Lead

    First action

    Draft the page outline with the operations team.

    How to measure it

    Data-collection inquiries

    Now
    –
    Target
    10 a month
    By
    90 days

    Steps

    1. 1

      Outline

      Modalities, scale, quality, scoping.

      Marketing Lead 2 days

    2. 2

      Check claims

      With operations and research.

      Member of Technical Staff, Research Engineer 3 days

    3. 3

      Write and design

      One page with a sample link.

      Marketing Lead 1 week

    4. 4

      Publish

      Menu and sitemap.

      Product Engineer 1 day

    5. 5

      Link from posts

      ABC and research posts.

      Marketing Lead 1 day

    6. 6

      Track

      Inquiries by source.

      Marketing Lead monthly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Marketing Lead

    Smallest test

    Publish a short version with a contact form.

    Why it works

    Write /products/data: modalities (teleoperation, egocentric, DAgger), embodiments and stations, scale and turnaround, quality checks (synchronization, alignment, audit), how scoping works, and a link to ABC-130K as a public sample. Add it to the menu and sitemap.

    Labs searching for robot training data need a page to land on; without it, rivals with data pages win the search.

    Risks and how to handle them

    • Customers want discretion

      Describe capabilities, not customer names.

    Evidence

    • Your websiteSitemap (October 22, 2026): 36 URLs, of which 26 are job posts, four are blog posts and one is a product page (/products/evaluations); /about is not listed; there is no page for the data collection service, the datasets or data partners, and no page in Spanish, Portuguese or Indonesian.Source
    • HiringXDOF's description in its job posts: "Frontier labs are racing to build general-purpose robots, and high-quality training data is the bottleneck. We're building the foundation behind the foundation models - the data collection systems, operational capability, exabyte-scale data warehouse, and software toolchain". The Engagement Manager tracks "SLAs, quality metrics, revenue consumption, and overall account health".Source
    • Your websiteAbout page: XDOF partners "with the world's leading robotics labs and most promising robotics companies to deliver production-scale datasets, advanced robotic systems, and the tooling"; "Our co-founders authored GELLO, now commonly deployed for teleoperation." The launch post adds a team from "Covariant, Tesla, Google, Uber, Applied Intuition" with "roots in Berkeley AI research".Source
    • Competitor researchRivals' own homepages (October 22, 2026): Scale AI, "Building the Data Foundry for Physical AI"; Encord, "the multimodal data layer for physical AI", "Trusted by 300+ AI teams"; Claru, "500K+ egocentric videos. 4M+ annotations. The only training data company 100% focused on physical AI."Source
  • 04

    The contact page says to get in touch "about data, evaluations, and robotics infrastructure", but the HTML shows only "Let's work together" and the form is drawn by script. Every inquiry arrives the same way, so the right person cannot answer fast.

    Medium impactLow effort First result in 1 week Marketing Lead

    First action

    Add the "I'm interested in" choice.

    How to measure it

    First reply

    Now
    –
    Target
    Under 1 business day
    By
    30 days

    Steps

    1. 1

      Add the choice

      Six options.

      Product Engineer 1 day

    2. 2

      Add fields

      Per option.

      Product Engineer 1 day

    3. 3

      Route

      To owners.

      Marketing Lead 1 day

    4. 4

      Set reply target

      One business day.

      Engagement Manager 1 day

    5. 5

      Server-render text

      Heading, options, intro.

      Product Engineer 1 day

    6. 6

      Review monthly

      Inquiries by type.

      Marketing Lead monthly

    First result
    1 week
    Cost
    $0
    Owner
    Marketing Lead

    Smallest test

    Add the choice and watch a month of inquiries.

    Why it works

    Add a required "I'm interested in" choice (evaluations, data collection, datasets, data venue partnership, press, other), a few fields per choice (robot, tasks, timeline), and route each to the right owner with a one-day reply target. Put the form text in the HTML.

    Fast, specific replies win early deals, and the routing data shows which offer draws demand.

    Risks and how to handle them

    • Longer form fewer submits

      Keep extra fields optional.

    Evidence

    • Your websiteContact page: description "Get in touch with XDOF about data, evaluations, and robotics infrastructure"; the HTML shows only the heading "Let's work together", with the form drawn by script.Source
    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
  • 05

    XDOF published four strong posts (ABC-130K, SARM2, noise floor) in two weeks in June and nothing since. Researchers follow labs that ship regularly, and the Marketing Lead owns "thought leadership".

    Medium impactLow effort First result in 1 month Marketing Lead

    First action

    List three posts that could ship from current work.

    How to measure it

    Email subscribers

    Now
    –
    Target
    1,000
    By
    2 quarters

    Steps

    1. 1

      List topics

      Three from current work.

      Marketing Lead 2 days

    2. 2

      Assign authors

      Researchers and ops.

      Marketing Lead 1 day

    3. 3

      Edit

      Short and clear.

      Marketing Lead monthly

    4. 4

      Add sign-up

      On the blog.

      Product Engineer 1 day

    5. 5

      Share

      LinkedIn and X.

      Marketing Lead monthly

    6. 6

      Review

      Visits and sign-ups.

      Marketing Lead quarterly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Marketing Lead

    Smallest test

    Publish one post this month.

    Why it works

    Set a monthly cadence: one research note, release update or field story. Draft from work already done (ABC-Eval progress, new tasks, collection lessons), repost to LinkedIn and X, and add an email sign-up.

    A steady cadence keeps the June audience and gives every launch a home.

    Risks and how to handle them

    • Researchers have no time

      Interview them and write it up.

    Evidence

    • Your websiteBlog: four posts, all published June 17-30, 2026 ("Announcing XDOF", "ABC-130K: The largest open source teleoperation dataset", "The Robot That Learned to Grade Itself" on the SARM2 reward model, and "Lowering the Noise Floor on Robot Data"); nothing since.Source
    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
    • Your websiteThe site links to LinkedIn (company/xdofai) and X (@xdofai) only; no YouTube channel or newsletter sign-up is linked.Source
  • 06

    The co-founders authored GELLO, "now commonly deployed for teleoperation", and XDOF worked on ABC and SARM2. That record is spread across the About page and blog posts; there is no single page a lab can scan.

    Medium impactLow effort First result in 2 weeks Marketing Lead

    First action

    List every paper and release.

    How to measure it

    Research-to-contact rate

    Now
    –
    Target
    5%
    By
    90 days

    Steps

    1. 1

      List work

      Papers, data, code.

      Member of Technical Staff, Research Engineer 2 days

    2. 2

      Write summaries

      One line each.

      Marketing Lead 2 days

    3. 3

      Link services

      Evaluations or data.

      Marketing Lead 1 day

    4. 4

      Publish

      Menu and sitemap.

      Product Engineer 1 day

    5. 5

      Link out

      Homepage and dataset card.

      Marketing Lead 1 day

    6. 6

      Keep current

      Each new paper.

      Member of Technical Staff, Research Engineer ongoing

    First result
    2 weeks
    Cost
    $0
    Owner
    Marketing Lead

    Smallest test

    Publish the page with three entries.

    Why it works

    Create /research with each project: one-line summary, paper link, code and data, and the XDOF service it relates to. Link it from the homepage and dataset card.

    Research credibility is how robotics teams choose data partners, and one page makes it easy to share.

    Risks and how to handle them

    • Shared credit with academic partners

      Name the collaborating labs on each entry.

    Evidence

    • Your websiteAbout page: XDOF partners "with the world's leading robotics labs and most promising robotics companies to deliver production-scale datasets, advanced robotic systems, and the tooling"; "Our co-founders authored GELLO, now commonly deployed for teleoperation." The launch post adds a team from "Covariant, Tesla, Google, Uber, Applied Intuition" with "roots in Berkeley AI research".Source
    • Your websiteBlog: four posts, all published June 17-30, 2026 ("Announcing XDOF", "ABC-130K: The largest open source teleoperation dataset", "The Robot That Learned to Grade Itself" on the SARM2 reward model, and "Lowering the Noise Floor on Robot Data"); nothing since.Source
    • Your websiteABC-130K post: "over 130,000 episodes across 195 bimanual manipulation tasks" under Apache 2.0; ABC-Sim and ABC-Eval marked "(coming soon)"; episodes viewable on Foundry; built with the UC Berkeley authors of "Scalable Behavior Cloning with Open Data, Training, and Evaluation"; "if you ever need more data or evaluations like this, we'd love to talk".Source

This quarter 9

The core plays for the next three months.

  • 07

    ABC-Eval is marked "coming soon", XDOF already ran real and simulated evaluations for the ABC authors, and the new Robotics Platform Engineer will build "partner-facing reporting" for evaluation runs. A public leaderboard on fixed physical tasks would be rare in robotics.

    High impactHigh effort First result in 2 quarters Robotics Platform Engineer

    First action

    Pick the first three tasks and the rules.

    How to measure it

    Entries

    Now
    –
    Target
    30
    By
    2 quarters

    Steps

    1. 1

      Pick tasks

      Three to five.

      Member of Technical Staff, Research Engineer 2 weeks

    2. 2

      Write rules

      Submission and scoring.

      Robotics Platform Engineer 2 weeks

    3. 3

      Build the page

      Results and videos.

      Robotics Platform Engineer 1 month

    4. 4

      Seed entries

      Baselines from ABC.

      Member of Technical Staff, Research Engineer 1 month

    5. 5

      Launch

      With the dataset audience.

      Marketing Lead 1 week

    6. 6

      Convert

      Offer private runs.

      Engagement Manager ongoing

    First result
    2 quarters
    Cost
    Station time
    Owner
    Robotics Platform Engineer

    Smallest test

    Run one task with five invited labs.

    Why it works

    Launch a leaderboard on three to five ABC tasks: teams submit a checkpoint, XDOF runs it on YAM stations, and results with episode videos are published. Offer free entries to academic labs and paid private runs for companies.

    Leaderboards bring repeat visits and citations, and every entry is a hands-on trial of the Evaluations service.

    Risks and how to handle them

    • Station capacity

      Cap free entries per month.

    Evidence

    • Your websiteABC-130K post: "over 130,000 episodes across 195 bimanual manipulation tasks" under Apache 2.0; ABC-Sim and ABC-Eval marked "(coming soon)"; episodes viewable on Foundry; built with the UC Berkeley authors of "Scalable Behavior Cloning with Open Data, Training, and Evaluation"; "if you ever need more data or evaluations like this, we'd love to talk".Source
    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
    • HiringProduct hiring: a Robotics Platform Engineer (posted October 22, 2026) to build "the product that lets partners configure, launch, and monitor evaluation runs", with "partner-facing reporting"; a Product Engineer owns the data and research platform, "the interface through which researchers at top frontier labs explore, purchase, and work with XDOF's data", including "self serve data acquisition flows for researchers".Source
    • Hugging FaceXDOF/ABC-130k on Hugging Face (October 22, 2026): 234,346 downloads in the last month and 1,924,086 in total, 135 likes, 65 followers of the XDOF organization and 10 discussion threads. Access is gated with automatic approval and shares each requester's username and email with XDOF. The dataset card points to the paper site and code but not to xdof.ai or XDOF Evaluations.Source
  • 08

    XDOF is hiring to build the product that lets partners "configure, launch, and monitor evaluation runs". Labs that want results in 24 hours should be able to join before it ships.

    High impactHigh effort First result in 2 quarters Robotics Platform Engineer

    First action

    Add the waitlist form.

    How to measure it

    Qualified waitlist

    Now
    –
    Target
    50 teams
    By
    1 quarter

    Steps

    1. 1

      Add the form

      Evaluations page.

      Product Engineer 2 days

    2. 2

      Pick partners

      Five design partners.

      Engagement Manager 1 month

    3. 3

      Build

      Configure, launch, monitor.

      Robotics Platform Engineer 1 quarter

    4. 4

      Launch content

      Post and video.

      Marketing Lead 1 month

    5. 5

      Pricing guide

      Per run or per hour.

      Co-founders 2 weeks

    6. 6

      Onboard

      From the waitlist.

      Engagement Manager ongoing

    First result
    2 quarters
    Cost
    Engineering time
    Owner
    Robotics Platform Engineer

    Smallest test

    Open the waitlist this month.

    Why it works

    Open a waitlist on the Evaluations page with robot, task and volume fields, invite the first design partners from the list, and launch with a post, demo video and pricing guidance.

    A waitlist sizes demand and lines up launch customers, and the launch becomes the first big story for the Marketing Lead.

    Risks and how to handle them

    • Delays

      Keep the waitlist updated monthly.

    Evidence

    • HiringProduct hiring: a Robotics Platform Engineer (posted October 22, 2026) to build "the product that lets partners configure, launch, and monitor evaluation runs", with "partner-facing reporting"; a Product Engineer owns the data and research platform, "the interface through which researchers at top frontier labs explore, purchase, and work with XDOF's data", including "self serve data acquisition flows for researchers".Source
    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
  • 09

    datasets.xdof.ai is where researchers will "explore, purchase, and work with XDOF's data", and the ABC-130K post links to it, but its page title names neither XDOF nor datasets and its HTML has no text or description. Links shared in chats or papers show nothing useful.

    Medium impactMedium effort First result in 1 quarter Product Engineer

    First action

    Change the page title and description.

    How to measure it

    Catalog visits

    Now
    –
    Target
    +50%
    By
    1 quarter

    Steps

    1. 1

      Title and description

      Say what it is.

      Product Engineer 1 hour

    2. 2

      Preview image

      For shares.

      Marketing Lead 1 day

    3. 3

      Catalog intro

      Server-rendered.

      Product Engineer 1 week

    4. 4

      Dataset entries

      One per dataset.

      Product Engineer 2 weeks

    5. 5

      Cross-link

      Main site and Evaluations.

      Product Engineer 1 day

    6. 6

      Measure

      Visits and sources.

      Marketing Lead monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Product Engineer

    Smallest test

    Fix the title and preview this week.

    Why it works

    Give it a proper title, description and preview image, server-render a short catalog intro and one entry per dataset, and link back to xdof.ai and Evaluations.

    A clear catalog page turns shared links into visits and makes the datasets visible to search and AI assistants.

    Risks and how to handle them

    • Private datasets exposed

      List only public releases.

    Evidence

    • Your websitedatasets.xdof.ai, linked from the ABC-130K post, serves a page whose one-word title names neither XDOF nor datasets, with no text or description in its HTML.Source
    • HiringProduct hiring: a Robotics Platform Engineer (posted October 22, 2026) to build "the product that lets partners configure, launch, and monitor evaluation runs", with "partner-facing reporting"; a Product Engineer owns the data and research platform, "the interface through which researchers at top frontier labs explore, purchase, and work with XDOF's data", including "self serve data acquisition flows for researchers".Source
    • Your websiteABC-130K post: "over 130,000 episodes across 195 bimanual manipulation tasks" under Apache 2.0; ABC-Sim and ABC-Eval marked "(coming soon)"; episodes viewable on Foundry; built with the UC Berkeley authors of "Scalable Behavior Cloning with Open Data, Training, and Evaluation"; "if you ever need more data or evaluations like this, we'd love to talk".Source
  • 10

    The Marketing Lead is hired for "Real events experience" and to turn "a conference or a dinner into pipeline". XDOF's product is physical: a policy running on a YAM station is more convincing than a booth.

    High impactMedium effort First result in 2 quarters Marketing Lead

    First action

    Pick the next two conferences.

    How to measure it

    Trials booked per event

    Now
    –
    Target
    5
    By
    2 quarters

    Steps

    1. 1

      Pick events

      Two next conferences.

      Marketing Lead 1 week

    2. 2

      Plan the station

      Portable YAM setup.

      Member of Technical Staff, Research Engineer 1 month

    3. 3

      Invite

      15 lab leads per dinner.

      Co-founders 1 month

    4. 4

      Run

      Station and dinner.

      Marketing Lead per event

    5. 5

      Follow up

      Free first round.

      Engagement Manager 1 week

    6. 6

      Measure

      Trials and deals.

      Marketing Lead per event

    First result
    2 quarters
    Cost
    Travel and event costs
    Owner
    Marketing Lead

    Smallest test

    Host one dinner at the next conference.

    Why it works

    At the main robot learning conferences, run a live station where researchers watch their own checkpoint or a public one run, and host a dinner for 15 lab leads with a founder. Follow up with a free evaluation round.

    Researchers trust what they can watch, and small dinners build relationships booths cannot.

    Risks and how to handle them

    • Hardware risk on site

      Bring a spare and a recorded fallback.

    Evidence

    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
    • Your websiteAbout page: XDOF partners "with the world's leading robotics labs and most promising robotics companies to deliver production-scale datasets, advanced robotic systems, and the tooling"; "Our co-founders authored GELLO, now commonly deployed for teleoperation." The launch post adds a team from "Covariant, Tesla, Google, Uber, Applied Intuition" with "roots in Berkeley AI research".Source
  • 11

    XDOF is signing retail, logistics, industrial and urban venues in Mexico and Brazil, and deploying devices across retail locations in Indonesia. Venue owners will look XDOF up after the first meeting, and the site has no page for them in any language.

    High impactMedium effort First result in 1 quarter LATAM Operations Lead

    First action

    Draft the English version with the LATAM lead.

    How to measure it

    Venues signed

    Now
    –
    Target
    +30%
    By
    2 quarters

    Steps

    1. 1

      Draft English

      Offer, privacy, setup.

      LATAM Operations Lead 1 week

    2. 2

      Translate

      Spanish, Portuguese, Indonesian.

      Brand Marketing (Indonesia) 2 weeks

    3. 3

      Add forms

      Routed to local leads.

      Product Engineer 1 week

    4. 4

      Publish

      With hreflang.

      Product Engineer 1 day

    5. 5

      Use in outreach

      Every meeting.

      Retail Operations Manager ongoing

    6. 6

      Measure

      Signed venues.

      LATAM Operations Lead monthly

    First result
    1 quarter
    Cost
    Translation costs
    Owner
    LATAM Operations Lead

    Smallest test

    Use a one-page Spanish version in ten venue meetings.

    Why it works

    Write one partner page per language: what XDOF collects, what venues get (fees, schedules, privacy), how setup works, and a contact form that routes to the local lead. Use it in every outreach and leave-behind.

    A clear page in the venue owner's language answers trust questions before they stall a deal.

    Risks and how to handle them

    • Privacy concerns

      Explain consent and data handling plainly.

    Evidence

    • HiringLATAM Operations Lead (Mexico or Brazil, posted September 20, 2026): builds "a diversified network of data collection venues and access points (retail, logistics, industrial, urban, etc.)", controls "cost per usable hour", and Spanish or Portuguese is "highly preferred".Source
    • HiringIndonesia hiring in Jakarta: Brand Marketing (posted October 22, 2026) for "XDOF AI Indonesia" to manage "our website and social media channels, including TikTok and Instagram", work with creators and "Develop SEO keywords"; a Retail Operations Manager (October 14) to coordinate "robotic devices across retail locations"; and a Project Manager (October 16) for "large-scale collection initiatives" across acquisition, implementation, audit and annotation teams.Source
    • Your websiteSitemap (October 22, 2026): 36 URLs, of which 26 are job posts, four are blog posts and one is a product page (/products/evaluations); /about is not listed; there is no page for the data collection service, the datasets or data partners, and no page in Spanish, Portuguese or Indonesian.Source
    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
  • 12

    The Jakarta brand role will shoot short videos, run TikTok and Instagram and work with creators. Robots learning to fold and sort in real shops are naturally watchable and explain what XDOF does to local partners and recruits.

    Medium impactLow effort First result in 1 quarter Brand Marketing

    First action

    Shoot the first three clips.

    How to measure it

    Social-sourced inquiries

    Now
    –
    Target
    20 a month
    By
    2 quarters

    Steps

    1. 1

      Plan formats

      Three formats.

      Brand Marketing (Indonesia) 1 week

    2. 2

      Get permissions

      Venues and people filmed.

      Retail Operations Manager 1 week

    3. 3

      Shoot

      Three a week.

      Brand Marketing (Indonesia) weekly

    4. 4

      Post

      TikTok and Instagram.

      Brand Marketing (Indonesia) weekly

    5. 5

      Work with creators

      Two local creators.

      Brand Marketing (Indonesia) 1 month

    6. 6

      Report

      Reach and inquiries.

      Brand Marketing (Indonesia) monthly

    First result
    1 quarter
    Cost
    Small creator budget
    Owner
    Brand Marketing (Indonesia)

    Smallest test

    Post one week of clips.

    Why it works

    Post three short videos a week: a task being collected, a day in operations, a partner venue (with permission). Measure reach and inquiries from venues and applicants.

    Local visibility makes venue outreach and hiring easier, at very low cost.

    Risks and how to handle them

    • Filming shows customer data

      Review each clip before posting.

    Evidence

    • HiringIndonesia hiring in Jakarta: Brand Marketing (posted October 22, 2026) for "XDOF AI Indonesia" to manage "our website and social media channels, including TikTok and Instagram", work with creators and "Develop SEO keywords"; a Retail Operations Manager (October 14) to coordinate "robotic devices across retail locations"; and a Project Manager (October 16) for "large-scale collection initiatives" across acquisition, implementation, audit and annotation teams.Source
    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
  • 13

    The Marketing Lead may own "a technical report in partnership with our academic collaborators", and XDOF's posts already found that more demonstrations can make a robot worse. A report on data quality would set the terms buyers use.

    High impactMedium effort First result in 2 quarters Marketing Lead

    First action

    Agree the questions with one partner lab.

    How to measure it

    Inquiries citing the report

    Now
    –
    Target
    20
    By
    2 quarters

    Steps

    1. 1

      Pick questions

      Three data-quality questions.

      Member of Technical Staff, Research Engineer 2 weeks

    2. 2

      Run experiments

      With partners.

      Member of Technical Staff, Research Engineer 1 quarter

    3. 3

      Write

      Report and summary.

      Marketing Lead 1 month

    4. 4

      Release data

      Evaluation episodes.

      Member of Technical Staff, Research Engineer 2 weeks

    5. 5

      Launch

      Webinar and talks.

      Marketing Lead 1 month

    6. 6

      Follow up

      Labs that engage.

      Engagement Manager ongoing

    First result
    2 quarters
    Cost
    Research time
    Owner
    Marketing Lead

    Smallest test

    Publish one chart from the noise-floor work.

    Why it works

    Publish a report with partner labs: how noise, task mix and operator skill affect policy results, with released evaluation data. Launch with a webinar and conference talks.

    Defining what good data means positions XDOF as the standard, ahead of larger data vendors.

    Risks and how to handle them

    • Results unflattering

      Publish anyway; honesty builds trust.

    Evidence

    • HiringMarketing Lead (founding, San Mateo, posted September 24, 2026): owns "messaging, positioning, content, website, announcements, thought leadership"; runs "major product launches" and events; must "Reach our key audiences deliberately: customers, consortium and data partners, and prospective hires. Build a distinct message and engagement strategy for each"; may own "a technical report in partnership with our academic collaborators" and should know how "to turn a conference or a dinner into pipeline".Source
    • Your websiteBlog: four posts, all published June 17-30, 2026 ("Announcing XDOF", "ABC-130K: The largest open source teleoperation dataset", "The Robot That Learned to Grade Itself" on the SARM2 reward model, and "Lowering the Noise Floor on Robot Data"); nothing since.Source
    • Your websiteABC-130K post: "over 130,000 episodes across 195 bimanual manipulation tasks" under Apache 2.0; ABC-Sim and ABC-Eval marked "(coming soon)"; episodes viewable on Foundry; built with the UC Berkeley authors of "Scalable Behavior Cloning with Open Data, Training, and Evaluation"; "if you ever need more data or evaluations like this, we'd love to talk".Source
    • Competitor researchRivals' own homepages (October 22, 2026): Scale AI, "Building the Data Foundry for Physical AI"; Encord, "the multimodal data layer for physical AI", "Trusted by 300+ AI teams"; Claru, "500K+ egocentric videos. 4M+ annotations. The only training data company 100% focused on physical AI."Source
  • 14

    The Engagement Manager tracks SLAs, quality metrics and account health, and XDOF runs audit and annotation teams. Customers see deliveries, but rarely a clear summary of quality.

    Medium impactMedium effort First result in 1 quarter Engagement Manager

    First action

    Design the one-page report.

    How to measure it

    Accounts expanding

    Now
    –
    Target
    +20%
    By
    2 quarters

    Steps

    1. 1

      Design

      One page.

      Engagement Manager 1 week

    2. 2

      Automate

      From quality data.

      Product Engineer 1 month

    3. 3

      Send

      Monthly.

      Engagement Manager monthly

    4. 4

      Collect feedback

      From customers.

      Engagement Manager monthly

    5. 5

      Public page

      Anonymized figures.

      Marketing Lead 1 quarter

    6. 6

      Review

      Renewals.

      Engagement Manager quarterly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Engagement Manager

    Smallest test

    Send one report to one customer.

    Why it works

    Send each customer a monthly one-page report: hours delivered, quality checks passed, issues fixed and turnaround. With permission, turn anonymized figures into a public quality page.

    Visible quality reduces churn and gives sales measured proof without naming customers.

    Risks and how to handle them

    • Numbers reveal weak months

      Show trends and fixes.

    Evidence

    • HiringXDOF's description in its job posts: "Frontier labs are racing to build general-purpose robots, and high-quality training data is the bottleneck. We're building the foundation behind the foundation models - the data collection systems, operational capability, exabyte-scale data warehouse, and software toolchain". The Engagement Manager tracks "SLAs, quality metrics, revenue consumption, and overall account health".Source
    • HiringIndonesia hiring in Jakarta: Brand Marketing (posted October 22, 2026) for "XDOF AI Indonesia" to manage "our website and social media channels, including TikTok and Instagram", work with creators and "Develop SEO keywords"; a Retail Operations Manager (October 14) to coordinate "robotic devices across retail locations"; and a Project Manager (October 16) for "large-scale collection initiatives" across acquisition, implementation, audit and annotation teams.Source
    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
  • 15

    Evaluations run only on YAM arms, and the page says "If you're interested in other embodiments, we'd love to hear from you". The Robotics Platform Engineer will build evaluation frameworks; demand data should choose the next robot.

    Medium impactLow effort First result in 1 quarter Robotics Platform Engineer

    First action

    Add the embodiment form.

    How to measure it

    Requests logged

    Now
    –
    Target
    30
    By
    1 quarter

    Steps

    1. 1

      Add form

      Robot, tasks, volume.

      Product Engineer 1 day

    2. 2

      Promote

      Blog and dataset card.

      Marketing Lead 1 week

    3. 3

      Review

      Monthly.

      Robotics Platform Engineer monthly

    4. 4

      Pick next

      Top demand.

      Co-founders 1 quarter

    5. 5

      Announce

      Coming next list.

      Marketing Lead 1 week

    6. 6

      Pre-sell

      To requesters.

      Engagement Manager ongoing

    First result
    1 quarter
    Cost
    $0
    Owner
    Robotics Platform Engineer

    Smallest test

    Add the form this week.

    Why it works

    Add a short form for the robot, tasks and volume a team wants, and publish a "coming next" list once demand is clear.

    Building for proven demand avoids buying hardware no one books.

    Risks and how to handle them

    • Requests are not commitments

      Ask for a pilot deposit.

    Evidence

    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
    • HiringProduct hiring: a Robotics Platform Engineer (posted October 22, 2026) to build "the product that lets partners configure, launch, and monitor evaluation runs", with "partner-facing reporting"; a Product Engineer owns the data and research platform, "the interface through which researchers at top frontier labs explore, purchase, and work with XDOF's data", including "self serve data acquisition flows for researchers".Source

Big bets 3

Larger plays that can change the trajectory.

  • 16

    XDOF says its operations produced "100,000s of hours of robot data", and rivals such as Scale AI, Encord and Claru compete on scale claims. Aggregated findings on cost, quality and task difficulty would be original.

    High impactHigh effort First result in 4 quarters Member of Technical Staff

    First action

    Pick ten metrics that can be shared.

    How to measure it

    Downloads

    Now
    –
    Target
    5,000
    By
    1 year

    Steps

    1. 1

      Pick metrics

      Ten shareable.

      Member of Technical Staff, Research Engineer 1 month

    2. 2

      Aggregate

      Across projects.

      Member of Technical Staff, Research Engineer 2 months

    3. 3

      Review privacy

      No customer identifiable.

      Co-founders 1 month

    4. 4

      Write

      Report.

      Marketing Lead 1 month

    5. 5

      Launch

      Press and events.

      Marketing Lead 1 month

    6. 6

      Repeat

      Yearly.

      Marketing Lead yearly

    First result
    4 quarters
    Cost
    Research time
    Owner
    Member of Technical Staff, Research Engineer

    Smallest test

    Publish one metric in a blog post.

    Why it works

    Publish a yearly report: hours by task type, quality failure rates, time per usable hour, what drove policy gains. Keep customer data private and aggregated.

    Original numbers earn citations and press and make XDOF the reference in the category.

    Risks and how to handle them

    • Customers object

      Share only aggregated data and get sign-off.

    Evidence

    • Your websiteEvaluations page, the only product page: XDOF runs customers' policies "across our global robot fleet" on "the same operations infrastructure that produced 100,000s of hours of robot data"; evaluations run on "bimanual YAM arms from I2RT" ("If you're interested in other embodiments, we'd love to hear from you"); "We can turn around initial results within 24 hours"; DAgger-style data collection is offered; call to action "Talk to us".Source
    • HiringXDOF's description in its job posts: "Frontier labs are racing to build general-purpose robots, and high-quality training data is the bottleneck. We're building the foundation behind the foundation models - the data collection systems, operational capability, exabyte-scale data warehouse, and software toolchain". The Engagement Manager tracks "SLAs, quality metrics, revenue consumption, and overall account health".Source
    • Competitor researchRivals' own homepages (October 22, 2026): Scale AI, "Building the Data Foundry for Physical AI"; Encord, "the multimodal data layer for physical AI", "Trusted by 300+ AI teams"; Claru, "500K+ egocentric videos. 4M+ annotations. The only training data company 100% focused on physical AI."Source
    • HiringLATAM Operations Lead (Mexico or Brazil, posted September 20, 2026): builds "a diversified network of data collection venues and access points (retail, logistics, industrial, urban, etc.)", controls "cost per usable hour", and Spanish or Portuguese is "highly preferred".Source
  • 17

    The Product Engineer is hired for "self serve data acquisition flows for researchers". ABC-130K shows strong demand for open data; smaller labs need more of it but cannot run a custom project.

    High impactHigh effort First result in 3 quarters Product Engineer

    First action

    Price one task pack.

    How to measure it

    Task packs sold

    Now
    –
    Target
    20 a quarter
    By
    3 quarters

    Steps

    1. 1

      Define packs

      Tasks and sizes.

      Member of Technical Staff, Research Engineer 1 month

    2. 2

      Price

      Per pack.

      Co-founders 2 weeks

    3. 3

      Build checkout

      Datasets site.

      Product Engineer 1 quarter

    4. 4

      Plan capacity

      Station schedule.

      Project Manager (Indonesia) 1 month

    5. 5

      Launch

      To dataset users.

      Marketing Lead 1 month

    6. 6

      Measure

      Sales and repeat buyers.

      Product Engineer monthly

    First result
    3 quarters
    Cost
    Engineering time
    Owner
    Product Engineer

    Smallest test

    Sell one pack by hand to a dataset user.

    Why it works

    Offer task packs (for example 1,000 episodes of a chosen task on YAM stations) at a listed price with checkout on the datasets site, delivered in the same format as ABC-130K.

    Packaged data opens a self-serve revenue line for labs too small for custom work, and grows the install base for evaluations.

    Risks and how to handle them

    • Cannibalizes custom deals

      Keep packs to standard tasks.

    Evidence

    • HiringProduct hiring: a Robotics Platform Engineer (posted October 22, 2026) to build "the product that lets partners configure, launch, and monitor evaluation runs", with "partner-facing reporting"; a Product Engineer owns the data and research platform, "the interface through which researchers at top frontier labs explore, purchase, and work with XDOF's data", including "self serve data acquisition flows for researchers".Source
    • Hugging FaceXDOF/ABC-130k on Hugging Face (October 22, 2026): 234,346 downloads in the last month and 1,924,086 in total, 135 likes, 65 followers of the XDOF organization and 10 discussion threads. Access is gated with automatic approval and shares each requester's username and email with XDOF. The dataset card points to the paper site and code but not to xdof.ai or XDOF Evaluations.Source
    • Your websitedatasets.xdof.ai, linked from the ABC-130K post, serves a page whose one-word title names neither XDOF nor datasets, with no text or description in its HTML.Source
    • Your websiteABC-130K post: "over 130,000 episodes across 195 bimanual manipulation tasks" under Apache 2.0; ABC-Sim and ABC-Eval marked "(coming soon)"; episodes viewable on Foundry; built with the UC Berkeley authors of "Scalable Behavior Cloning with Open Data, Training, and Evaluation"; "if you ever need more data or evaluations like this, we'd love to talk".Source
  • 18

    XDOF has an active SAM.gov registration for all awards with research-and-development codes, expiring October 28, 2026. Government research programs fund robot learning and need data and evaluation partners.

    Medium impactMedium effort First result in 3 quarters Co founders

    First action

    Renew the registration.

    How to measure it

    Proposals

    Now
    –
    Target
    3
    By
    3 quarters

    Steps

    1. 1

      Renew

      Before expiry.

      Co-founders 1 week

    2. 2

      List programs

      Robot learning funding.

      Co-founders 1 month

    3. 3

      Find partners

      Academic labs.

      Member of Technical Staff, Research Engineer 1 month

    4. 4

      Write proposals

      Data and evaluation roles.

      Member of Technical Staff, Research Engineer 1 quarter

    5. 5

      Submit

      Three proposals.

      Co-founders 2 quarters

    6. 6

      Review

      Wins and lessons.

      Co-founders quarterly

    First result
    3 quarters
    Cost
    Proposal time
    Owner
    Co founders

    Smallest test

    Join one academic partner's proposal.

    Why it works

    Renew the registration, list target programs and academic partners already working with XDOF, and propose data and evaluation work as a subcontractor or partner on grants.

    Grants fund research the company wants to do anyway and add credibility with labs.

    Risks and how to handle them

    • Slow cycles

      Treat as a long-term channel.

    Evidence

    • Public recordsSAM.gov lists XDOF, INC as an active entity registration for all awards (registered September 26, 2025; expires October 28, 2026), with NAICS codes including 541715 (research and development in the physical, engineering and life sciences) and 541330 (engineering services).Source
    • Your websiteABC-130K post: "over 130,000 episodes across 195 bimanual manipulation tasks" under Apache 2.0; ABC-Sim and ABC-Eval marked "(coming soon)"; episodes viewable on Foundry; built with the UC Berkeley authors of "Scalable Behavior Cloning with Open Data, Training, and Evaluation"; "if you ever need more data or evaluations like this, we'd love to talk".Source
    • Your websiteAbout page: XDOF partners "with the world's leading robotics labs and most promising robotics companies to deliver production-scale datasets, advanced robotic systems, and the tooling"; "Our co-founders authored GELLO, now commonly deployed for teleoperation." The launch post adds a team from "Covariant, Tesla, Google, Uber, Applied Intuition" with "roots in Berkeley AI research".Source

Wildcards 3

Unexpected experiments, each with a cheap way to test it.

  • 19

    XDOF has 25 open roles, many in robotics research and engineering, and the people downloading ABC-130K are robotics researchers. The same audience feeds customers and hires.

    Medium impactLow effort First result in 1 quarter Technical Sourcer

    First action

    Add the hiring line to the card.

    How to measure it

    Applicants from Hugging Face

    Now
    –
    Target
    20 a quarter
    By
    2 quarters

    Steps

    1. 1

      Add line

      Card and landing page.

      Marketing Lead 1 day

    2. 2

      Plan a talk

      Research team.

      Member of Technical Staff, Research Engineer 2 weeks

    3. 3

      Invite

      Opted-in users.

      Technical Sourcer 1 week

    4. 4

      Run

      Online talk.

      Member of Technical Staff, Research Engineer 1 month

    5. 5

      Track

      By source.

      Technical Recruiter monthly

    6. 6

      Repeat

      Each release.

      Marketing Lead per release

    First result
    1 quarter
    Cost
    $0
    Owner
    Technical Sourcer

    Smallest test

    Add the line and count applicants.

    Why it works

    Add a "We're hiring robotics researchers" line to the dataset card and landing page, and invite opted-in requesters to an online talk about the work. Track applicants by source.

    Researchers who already use XDOF data understand the work and apply with context.

    Risks and how to handle them

    • Mixes customer and hiring messages

      Keep hiring to one line.

    Evidence

    • Hugging FaceXDOF/ABC-130k on Hugging Face (October 22, 2026): 234,346 downloads in the last month and 1,924,086 in total, 135 likes, 65 followers of the XDOF organization and 10 discussion threads. Access is gated with automatic approval and shares each requester's username and email with XDOF. The dataset card points to the paper site and code but not to xdof.ai or XDOF Evaluations.Source
    • Your websiteSitemap (October 22, 2026): 36 URLs, of which 26 are job posts, four are blog posts and one is a product page (/products/evaluations); /about is not listed; there is no page for the data collection service, the datasets or data partners, and no page in Spanish, Portuguese or Indonesian.Source
    • Your websiteHomepage (October 22, 2026): title "XDOF" and description "Defining motion for autonomous systems."; the HTML sent to browsers and crawlers holds only the menu and footer. The hero loads by script; its code holds one sentence ("we are a group of engineers and roboticists intensely focused on assembling the ingredients for the first generation of robotics foundation models"), "Backed by Thrive Capital, WndrCo, Andreessen Horowitz (a16z), and Samsung Next." and a "Join the team" panel whose button is "See open roles".Source
  • 20

    The co-founders authored GELLO, widely used for teleoperation, and the site links only to LinkedIn and X. Tutorials from the source would draw the builders who later need data and evaluations.

    Medium impactLow effort First result in 1 quarter Member of Technical Staff

    First action

    Record the first tutorial.

    How to measure it

    Subscribers

    Now
    –
    Target
    2,000
    By
    2 quarters

    Steps

    1. 1

      Plan episodes

      Five topics.

      Member of Technical Staff, Research Engineer 1 week

    2. 2

      Record

      One a month.

      Member of Technical Staff, Research Engineer monthly

    3. 3

      Edit

      Short and clear.

      Marketing Lead monthly

    4. 4

      Publish

      YouTube and site.

      Marketing Lead monthly

    5. 5

      Link services

      End screens.

      Marketing Lead monthly

    6. 6

      Review

      Views and inquiries.

      Marketing Lead quarterly

    First result
    1 quarter
    Cost
    Recording time
    Owner
    Member of Technical Staff, Research Engineer

    Smallest test

    Publish one tutorial.

    Why it works

    Record short videos on setting up teleoperation, collecting clean demonstrations and avoiding the noise problems from the blog. Start a YouTube channel and link it from the site.

    Teaching the method builds trust and an audience that becomes customers and hires.

    Risks and how to handle them

    • Time from researchers

      Short recordings, edited by marketing.

    Evidence

    • Your websiteAbout page: XDOF partners "with the world's leading robotics labs and most promising robotics companies to deliver production-scale datasets, advanced robotic systems, and the tooling"; "Our co-founders authored GELLO, now commonly deployed for teleoperation." The launch post adds a team from "Covariant, Tesla, Google, Uber, Applied Intuition" with "roots in Berkeley AI research".Source
    • Your websiteThe site links to LinkedIn (company/xdofai) and X (@xdofai) only; no YouTube channel or newsletter sign-up is linked.Source
    • Your websiteBlog: four posts, all published June 17-30, 2026 ("Announcing XDOF", "ABC-130K: The largest open source teleoperation dataset", "The Robot That Learned to Grade Itself" on the SARM2 reward model, and "Lowering the Noise Floor on Robot Data"); nothing since.Source
  • 21

    The LATAM lead manages "cost per usable hour", which is also how labs think about data. Rivals publish scale claims; few explain cost.

    Medium impactMedium effort First result in 2 quarters LATAM Operations Lead

    First action

    Agree the ranges with the founders.

    How to measure it

    Days to signed project

    Now
    –
    Target
    -20%
    By
    2 quarters

    Steps

    1. 1

      Gather data

      Recent projects.

      LATAM Operations Lead 2 weeks

    2. 2

      Set ranges

      Typical projects.

      Co-founders 2 weeks

    3. 3

      Write guide

      Short and clear.

      Marketing Lead 1 week

    4. 4

      Review

      Founders.

      Co-founders 1 week

    5. 5

      Publish

      Data page.

      Product Engineer 1 day

    6. 6

      Use in sales

      Every call.

      Engagement Manager ongoing

    First result
    2 quarters
    Cost
    Staff time only
    Owner
    LATAM Operations Lead

    Smallest test

    Share the guide with three prospects.

    Why it works

    Publish a short guide: what drives cost per usable hour (task, embodiment, location, quality level), ranges for typical projects and how XDOF scopes them. Use it in sales calls.

    Price clarity shortens procurement and filters out poor-fit leads.

    Risks and how to handle them

    • Locks in prices

      Publish ranges with clear drivers.

    Evidence

    • HiringLATAM Operations Lead (Mexico or Brazil, posted September 20, 2026): builds "a diversified network of data collection venues and access points (retail, logistics, industrial, urban, etc.)", controls "cost per usable hour", and Spanish or Portuguese is "highly preferred".Source
    • HiringXDOF's description in its job posts: "Frontier labs are racing to build general-purpose robots, and high-quality training data is the bottleneck. We're building the foundation behind the foundation models - the data collection systems, operational capability, exabyte-scale data warehouse, and software toolchain". The Engagement Manager tracks "SLAs, quality metrics, revenue consumption, and overall account health".Source
    • Competitor researchRivals' own homepages (October 22, 2026): Scale AI, "Building the Data Foundry for Physical AI"; Encord, "the multimodal data layer for physical AI", "Trusted by 300+ AI teams"; Claru, "500K+ egocentric videos. 4M+ annotations. The only training data company 100% focused on physical AI."Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

ProgrammableTalk

Fill the events in this plan

These ideas are events. Talk gets registrants for them and tells you who's coming and what their company does before the day.

Plan an event (opens programmable.talk in a new tab)
ProgrammableDigital

Show it, don't just describe it

These ideas need something to show: a demo, a video, an animation, graphics or a deck. Digital turns screenshots or the site into all of them, in the company's own brand.

Make a demo (opens programmable.digital in a new tab)
More example playbooks and reviews

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • 3xvoice growth strategyBuilds custom AI voice receptionists and automations for service businesses: an agent scripted for each client's services and pricing…22 growth ideas
  • General Astronautics growth strategyA Y Combinator (Winter 2026) space robotics startup building autonomous dexterous robotic arms for orbital inspection, repair and…23 growth ideas
  • BiggerPockets growth strategyThe largest real estate investing community and media network: forums, calculators and tools, a network of podcasts and YouTube shows…21 growth ideas
  • Perchwell growth strategyA modern MLS and listings platform for real estate agents, brokerages and MLS organizations: search, listing entry, client collaboration…21 growth ideas
  • SNR Growth growth strategyA Sydney-based growth agency for marketplace startups. It embeds specialists who run growth audits and strategy, fractional growth…23 growth ideas
  • Mineral Project growth strategyAn Indian home fragrance brand selling hand-poured soy candles, room and linen sprays, incense and a small body line, all through its own…21 growth ideas
  • Ordinary Habit growth strategyA design-led puzzle brand selling 100, 500 and 1,000-piece jigsaw puzzles, matching cards, journals and home goods with original art by…21 growth ideas
  • Orium growth strategyA composable commerce and digital experience consultancy and systems integrator (formerly Myplanet) that designs and builds enterprise…22 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on October 23, 2026, and cites them throughout. XDOF did not commission, review or endorse it, and Programmable is not affiliated with XDOF. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how XDOF runs its business.

Something wrong, or do you represent XDOF? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 306 ideas

    Dataset card and requester outreach, homepage paths, data service page, contact routing, blog restart and research page.

  2. Days 31 to 6012 ideas

    ABC-Eval leaderboard, self-serve eval waitlist, datasets site, conferences, local partner pages, Indonesia social, the technical report, quality reports and embodiment demand; dataset-user hiring, tutorials and the pricing guide.

  3. Days 61 to 903 ideas

    Commit to the annual data report, self-serve task packs and research awards.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value high-quality, large-scale robot data. The most cited problem: training data is the bottleneck.

  • Fast real-world evaluations
  • Running evaluations in-house is slow
  • In their words: “teleoperation”

Market

Competitive landscape

3 competitors compared: Scale AI, Encord and Claru.

  • Gap to exploit: Open real-robot evaluation benchmarks
  • Scale AI: General data vendor, less robotics research depth
  • Encord: Software, not collection operations

Hiring signals

XDOF is building its first marketing function and its first self-serve products.

  • Marketing Lead: Founding marketer for customers, partners and hires. Ideas 1-6, 10, 13.
  • Robotics Platform Engineer: Partner product for evaluation runs. Ideas 7, 8, 15.
  • LATAM Operations Lead: Data venues in Mexico and Brazil; cost per usable hour. Ideas 11, 21.

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks firstNew accounts get 100,000 free enrichment credits with code BETA100K (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammableMediaAn AI note taker, interpreter and webinar host for your calls (opens programmable.media in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence