Base

Booster

Axis Robotics (<a style="text-decoration:none" href="https://x.com/axisrobotics" rel="noopener noreferrer nofollow">@axisrobotics</a>) is a San Francisco startup that crowdsources training data for robots through browser simulations and a mobile app, recording verified contributions on Base. On July 27, it announced a $12 million seed round led by Hack VC (<a style="text-decoration:none" href="https://x.com/hack_vc" rel="noopener noreferrer nofollow">@hack_vc</a>), with participation from Nomad Capital (<a style="text-decoration:none" href="https://x.com/NomadCapital_io" rel="noopener noreferrer nofollow">@NomadCapital_io</a>), Pi Network Ventures (<a style="text-decoration:none" href="https://x.com/PiCoreTeam" rel="noopener noreferrer nofollow">@PiCoreTeam</a>), 10K Ventures (<a style="text-decoration:none" href="https://x.com/10kventure" rel="noopener noreferrer nofollow">@10kventure</a>), and angel investors.
For Pi Network watchers, the deal extends a pattern. The ecosystem's $100 million venture arm has now placed a second disclosed bet on the robotics and AI stack, after joining OpenMind's $20 million round led by Pantera Capital in 2025. Add the strategic investment in CiDi Games, and a picture emerges of a fund looking well beyond crypto-native projects.
<h2 id="span-span-strong-what-is-axis-robotics-strong-span-span">What Is Axis Robotics?</h2>
Founded in 2025 and headquartered in San Francisco, Axis Robotics describes itself as "the compounding data engine accelerating Physical AI." The company is small, with roughly 11 employees at the time of the round, but the credentials run deep: AI and robotics researchers from UC Berkeley, Carnegie Mellon, Georgia Tech, NTU, and SJTU, alongside growth operators who have scaled consumer products past 30 million users.
Founder Chris Feng (<a style="text-decoration:none" href="https://x.com/chris_anm01" rel="noopener noreferrer nofollow">@chris_anm01</a>) previously served as COO of Chainbase and has a background in consulting and venture investing, including time at Bain &amp; Company.
The company's core thesis is a data problem. Large language models had decades of internet text to train on. Robots have no equivalent. Machines that perceive and act in the physical world face what Axis calls three barriers: severe data scarcity, a generalization gap, and embodiment fragmentation, meaning data collected on one robot body rarely transfers cleanly to another.
<h2 id="span-span-strong-how-does-the-data-engine-work-strong-span-span">How Does the Data Engine Work?</h2>
Axis builds a human-in-the-loop pipeline with four main components:
<ul>
 <li>Task Gen Engine: procedurally generates diverse robotic tasks by randomizing objects, layouts, visuals, and robot types.</li>
 <li>Browser-based teleoperation: contributors control simulated robots from a web page, no hardware needed, producing motion trajectories the company claims arrive at roughly 10x the throughput of traditional lab collection.</li>
 <li>Ego capture app: a mobile app uses real-time hand pose tracking to gather first-person, real-world data from a global workforce.</li>
 <li>Processing pipeline: automated cleaning, domain randomization, and language annotation turn raw contributions into model-ready datasets.</li>
</ul>
The approach is "Simulation First." Contributors teleoperate simulated arms through tasks like pick-and-place, the trajectories get cleaned and augmented in high-fidelity backends such as IsaacSim, and the results train Vision-Language-Action models and imitation learning policies that transfer to physical robots. When a model fails, human corrections feed the next training round, a loop the company frames as compounding intelligence rather than a static dataset.
The crypto layer sits underneath. Accepted trajectories receive unique Data IDs recorded on&nbsp;<a style="text-decoration:none" href="https://bsc.news/post/base-l2-coinbase" rel="noopener noreferrer nofollow">Base</a> for verifiable ownership and provenance, wrapped in a Train-to-Earn incentive model. Axis launched its main product on Base in March 2026 after two early testing rounds. The first, "The Little Prince's Rose," collected 10,000+ valid trajectories in 3 days and trained a policy that ran on a physical Franka arm, autonomously completing a flower-watering task in an early proof of sim-to-real transfer. A product beta followed, drawing 20,000+ users who produced roughly 180,000 trajectories across 47 task types in 10 days. The company was also a finalist in Base Batches 003 and part of Base Founders Residency Batch 002.
<h2 id="span-span-strong-what-traction-does-axis-claim-strong-span-span">What Traction Does Axis Claim?</h2>
The numbers in the funding materials are substantial for a 2025 startup:
<ul>
 <li>Over 100,000 active contributors, submitting 3 to 4 times daily on average</li>
 <li>More than 1,200 hours of simulation data and 20,000+ hours of real-world egocentric data generated monthly</li>
 <li>90+ task categories, 2,000+ digital assets, and 200,000+ verified trajectories accumulated on the platform</li>
</ul>
The benchmark claim stands out. Pretraining the π0.5 robotics model, built by Physical Intelligence and unrelated to Pi Network despite the shared symbol, on Axis's diversified Sim Dataset V1 improved overall success by 4.9 points on LIBERO-Plus. That beat a volume-matched RoboCasa365 baseline by 31.3 points, with gains in layout generalization, sensor-noise resilience, and robot-pose robustness.
On the commercial side, Axis sells customized Task Packages to robotics hardware makers, Physical AI model companies, and industrial automation firms. Named partners include Booster Robotics, Manycore Tech, Feagine Robotics, Dexmal, Lotus, Geely Auto, and SomaStacks, among others.
<h2 id="span-span-strong-what-comes-next-strong-span-span">What Comes Next?</h2>
The proceeds will expand procedural generation capabilities and scale the contributor network into new regions, with paid pilots and a $1 million+ ARR target on the near-term roadmap. Two dataset releases are already scheduled: Sim Dataset V2 in September and a DAgger post-training dataset in November.
Feng framed the race in data terms rather than model terms, arguing that competitiveness in Physical AI depends on how quickly training data can be accumulated and improved. In the funding announcement, he put it bluntly: "Physical AI demands billions of human-physical interaction motion trajectories."
Whether a crowdsourced network can reach that scale is the question the $12 million has to answer. But with Hack VC leading, Pi Network Ventures deepening its robotics exposure, and two dataset releases due before year-end, Axis has given the market concrete milestones to check its claims against.
Sources:
<ul>
 <li><a style="text-decoration:none" href="https://x.com/axisrobotics/status/2081711331791827387" rel="noopener noreferrer nofollow">Axis Robotics</a> Official announcement of the $12 million seed round on X.</li>
 <li><a style="text-decoration:none" href="https://www.cryptopolitan.com/axis-robotics-raised-12m-funding-to-build-the-compounding-data-engine-accelerating-physical-ai/" rel="noopener noreferrer nofollow">Cryptopolitan</a> Press release detailing the round, investors, and the data engine architecture.</li>
 <li><a style="text-decoration:none" href="https://en.bloomingbit.io/feed/news/117048" rel="noopener noreferrer nofollow">Bloomingbit</a> Coverage including the dataset roadmap and founder comments.</li>
 <li><a style="text-decoration:none" href="https://axisrobotics.ai" rel="noopener noreferrer nofollow">Axis Robotics website</a> Company site and documentation on the platform and metrics.</li>
</ul>

Pi Network Invests in Physical AI: What is Axis Robotics?

Table of Contents

BSCN

Other articles published on Jul 28, 2026

Join our free newsletter for daily crypto updates!