Michael Cho, founder of FrodoBots and Ultimate Fighting Bots, talks with The Humanoid Hub about robot entertainment, embodied AI research, and the BitRobot Network.
Cho says Ultimate Fighting Bots makes robot fights entertaining even while the robots are still clumsy. He sees robotic gaming and entertainment as a way to give people a reason to engage with robots as their capabilities improve. His interest in robots dates back to growing up with anime such as Gundam and watching BattleBots.
FrodoBots began as a weekend project to build remotely controlled sidewalk robots. Its cheapest robot costs about $200, and players can drive it over the internet in a game modeled partly on Pokémon Go. The company records video and player control actions as teleoperation data; Cho describes an open dataset of about 2,000 hours from around 20 cities and a planned 10,000-hour dataset collected across 30 or 40 cities. In competitions at IROS and ICRA, human players have outperformed research teams’ AI systems at navigating the game.
Cho identifies three useful data sources: real-world teleoperation, simulation including world models, and video without action labels. Each has trade-offs, and he says physical AI also faces a difficult evaluation problem because robots must be tested in the changing real world. Failures can reveal problems, but edge cases become harder to find as systems improve. He argues that progress will require diverse data and major algorithmic advances, while questioning whether today’s compute-intensive path is the only way to achieve physical intelligence.
Cho describes BitRobot as a crypto-based effort to direct resources such as teleoperators, robots, compute, deployment labor, and repairs to embodied AI researchers. He frames it as a coordination and incentive challenge, with a long-term ambition of mobilizing $1 billion to $10 billion in resources. He says researchers could form subnets for projects such as world models, teleoperation, or exoskeletons, and that the separate BitRobot Foundation is led by Jonathan Victor and Amanda Young.
Cho says humanoids need both better hardware and better AI before they are suitable for homes. He questions whether joint-mounted actuators can be safe around people, and suggests tendon-like designs may be more compliant, though harder to manufacture and repair. He also points to the lack of human-quality touch sensing as a major limitation. He encourages researchers to seek resources through BitRobot and invites people to experiment with FrodoBots’ low-cost sidewalk robot.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
On Going Direct, host Jordan speaks with Chris Paxton, a roboticist at Agility Robotics, and Michael Cho, co-founder of Beat Robot and RoboPapers co-host, about robot learning, data, evaluation, and ways to get more robots into the world.
