Stanford ToddlerBot turns low-cost humanoid hardware into a learning platform

Stanford researchers Haochen Shi, Weizhuo Wang, Shuran Song, and C. Karen Liu submitted ToddlerBot in February 2025 as an open-source humanoid platform for policy learning and loco-manipulation. The robot is small by design: a 0.56-meter, 3.4-kilogram body built from 3D-printed parts and commercially available components, with total cost kept under $6,000.

The platform is aimed at the research bottlenecks that sit between simulation and full-size humanoid hardware. Large bipeds are expensive, hazardous, slow to repair, and hard for many labs to reproduce. ToddlerBot gives researchers a whole-body platform small enough for repeated experiments while still supporting walking, balance, perception, teleoperation, and manipulation learning.

The project materials describe simulation and real-world data collection, plug-and-play zero-point calibration, motor system identification for a digital twin, and teleoperation for demonstrations. A 2.0 release added crawling, faster walking, VR teleoperation, and more dynamic behaviors.

The paper covers parts availability, calibration, maintenance and transfer from simulation to hardware. The authors also report that an external team successfully replicated the robot.

Have a robotics update Korthos should review? Send news, deployments, product releases, funding rounds, research, or media to [email protected] or reach out on X at @agkorthos.