BitRobot

Training a large language model requires petabytes of human-generated text scraped from the internet. Training a robot to navigate a sidewalk, assemble furniture, or manipulate objects requires something the internet cannot provide: physical demonstrations in the real world. That data scarcity is the central bottleneck holding back embodied AI, and it is the problem BitRobot was built to solve.

Launched in February 2025 by FrodoBots Lab, BitRobot is a crypto-incentivized research network on Solana that organizes global contributors—hardware operators, gamers, university researchers, and compute providers—into coordinated missions that produce open-source robotics data and AI models. Its founding team pairs hardware-first experience with deep protocol expertise: Michael Cho co-founded FrodoBots Lab, Jonathan Victor previously led the Filecoin ecosystem at Protocol Labs, and Juan Benet is the founder of Protocol Labs itself.

The Subnet Model

BitRobot's architecture divides work into subnets—focused research environments, each designed around a specific data-collection or model-training objective. Subnet owners define the tasks and validation rules; subnet contributors supply robots, teleoperation labor, or compute; and subnet validators evaluate outputs. The network currently hosts ten active missions spanning a range of robotic embodiments and challenge types.

Several subnets are underway. ET Fugi (SN/01) deploys sidewalk rover robots that human operators guide through urban environments to catch simulated "alien fugitives," generating what the project describes as the largest sidewalk navigation dataset on Hugging Face. Seesaw (SN/02) collects egocentric video through mobile phones as a scalable proxy for robot-eye-view data, having logged more than 4.5 million completed tasks. TeleArms (SN/03) runs simulated robotic arm control exercises, while Axis Robotics (SN/04) uses teleop-in-simulation for manipulation tasks, collectively accumulating over 1,400 hours of recorded action data. Humanoids in the Wild (SN/05) collects real-world demonstrations using full humanoid robots—one of the most data-scarce domains in the field. RoboCap (SN/06) builds egocentric video datasets for the same purpose. Across all subnets, the network has completed more than 50,000 teleoperation missions.

Grand Challenges

Beyond data collection, BitRobot runs a competitive research grants program called the Grand Challenges. The BitRobot Foundation has pledged $5 million toward measurable robotics benchmarks with three active competitions. The Earth Rover Grand Challenge targets autonomous outdoor navigation. Robotic Origami benchmarks paper-folding precision against human expert performance. Robotic IKEA Assembly challenges AI systems to complete furniture assembly tasks end-to-end. These competitions are designed to produce openly shared results and advance the state of the art in ways any researcher can build on.

Who Is Building With BitRobot

BitRobot's research outputs have found traction in academic settings. A team at UC Berkeley trained a navigation model called LogoNav using crowd-sourced data from FrodoBots rover deployments, evaluating it on physical robots across six countries. Research citing the network's datasets has also come from MIT CSAIL, UCLA, Tampere University, Peking University, and the University of Amsterdam.

The network has drawn competitive teams from outside traditional research institutions as well. In 2025, eight university AI teams competed head-to-head against five human gamer teams from Yield Guild Games, completing identical navigation tasks across eight cities. The exercise produced both real-world training data and a direct comparison between human operator performance and current AI capabilities—with human operators outperforming AI systems under those conditions.

Incentive Architecture

BitRobot's economic design, detailed in its whitepaper, distinguishes between public subnets—which produce openly licensed datasets and models in exchange for network emissions—and private subnets, which retain proprietary ownership of their outputs. Individual robots can be registered as Embodied Node Tokens (ENTs), NFT-based digital identities that enable unique robot tracking and payment routing within the network.

Governance is organized through the BitRobot Foundation, which supports infrastructure and ecosystem grants, and the BitRobot Senate, a body of nominated representatives that votes on which subnets receive emission weight at regular voting epochs. A counterbalance mechanism called Gandalf AI—an open-source AI agent—can propose alternative weight distributions to check Senate concentration. Network participants can delegate voting power to either the Senate or Gandalf AI.

For data storage at the scale the network generates, BitRobot integrates with Filecoin, drawing on Protocol Labs' existing distributed storage infrastructure.

Funding and Backers

In February 2025, FrodoBots Lab disclosed $8 million in total funding to support the BitRobot launch and early network development. The seed round was led by Protocol VC, with participation from Big Brain Holdings, Fabric Ventures, Zee Prime Capital, Tioga Capital, Sfermion, Solana Ventures, and Virtuals Protocol. Angel investors included Solana Labs co-founders Anatoly Yakovenko and Raj Gokal, alongside founders from several DePIN projects.

"We need to incentivize a diverse set of stakeholders to dramatically accelerate robotics AI," Juan Benet said at the time of the announcement. The round reflected a thesis that the largest frontier in AI—getting robots to operate reliably in the physical world—requires the same kind of open, distributed infrastructure that Protocol Labs built for decentralized storage.

Early Trajectory

BitRobot entered 2026 with active subnets, a functioning Grand Challenges program, and demonstrated academic uptake of its research outputs. Its network-native token had not launched as of the project's 2025 recap, with the team prioritizing infrastructure, research programs, and contributor participation before opening market-side mechanics. The project positions itself as a long-horizon bet: building the open data commons that robotics researchers currently lack, one competitive subnet at a time.