ToddlerBot

Stanford's open-source toddler-sized humanoid robot, fully 3D-printed and built for machine-learning research.

ToddlerBot is an open-source small humanoid robot from Stanford researchers Haochen Shi, Weizhuo Wang, Shuran Song, C. Karen Liu, and others, published at CoRL 2025. It stands about 0.56 m tall and weighs about 3.4 kg in the paper's version (later versions are slightly heavier, around 3.7 kg), with 30 actively driven degrees of freedom (7 per arm, 6 per leg, and 2 each in the neck and waist). The entire robot is 3D-printed from off-the-shelf parts, keeping total cost under $6,000, and the team released both the code and the assembly documentation. It's meant to solve the problem that humanoid robots are usually too expensive and too fragile to use for large-scale data collection and experimentation: because a ToddlerBot is cheap and quick to repair after a fall, researchers can freely run reinforcement-learning walking, collect data via VR teleoperation, and train diffusion policies for whole-body tasks like grasping and pushing boxes; it's also commonly used to validate sim-to-real transfer.

ExampleIn the paper, demonstration data is collected through VR teleoperation, then used to train a diffusion policy so ToddlerBot can perform whole-body manipulation tasks like grasping and pushing.

Related

Small-size Humanoid Robot · Open-Source Hardware (OSHW) · Humanoid Robot · Sim-to-Real Transfer · Diffusion Policy · VR Teleoperation

Sources

ToddlerBot 项目主页 (Chinese)
ToddlerBot: Open-Source ML-Compatible Humanoid Platform for Loco-Manipulation (arXiv)

As of

2025-09