Toddlerbot: Open-Source Humanoid Robot (toddlerbot.github.io)

ToddlerBot is an open-source, low-cost humanoid robot platform designed to advance scalable policy learning and research in robotics and AI. Featuring 30 active degrees of freedom—including 7 DoFs per arm and 6 per leg—ToddlerBot combines sophisticated mechatronics with integrated sensors such as fisheye stereo cameras, an IMU, microphones, and a Jetson Orin NX onboard computer for real-time depth estimation. Its robust construction supports dynamic behaviors like cartwheels, crawling, and omnidirectional walking (up to 0.25 m/s speed) alongside manipulation skills such as bimanual grasping and pushing, all enabled through reinforcement learning (RL) and diffusion-based policies trained with demonstration data. Significantly, ToddlerBot’s design emphasizes durability and ease of repair, capable of withstanding multiple falls and quick restoration via 3D-printing and rapid assembly. The platform demonstrates advanced sim-to-real transfer for locomotion and manipulation, including zero-shot execution of complex tasks like pull-ups and push-ups, leveraging integrated real-time speech interfaces powered by OpenAI’s GPT-4o. Collaboration between multiple ToddlerBot instances shows promise for coordinated, multi-robot task execution. ToddlerBot’s fully open-source manuals and modular hardware make it an accessible tool for researchers and educators, potentially accelerating innovation in humanoid robotics, embodied AI, and multi-agent systems.