ToddlerBot: Open-Source ML-Compatible Humanoid Platform for Loco-Manipulation
TL;DR An open-source, fully 3D-printed miniature humanoid (0.56 m, 3.4 kg, 30 active DoFs) that costs under 6,000 USD and is built for collecting both simulation and real-world data. Diffusion policies trained on 60 teleoperated demonstrations reach 90% success on bimanual and 75% on full-body manipulation.
What they do
ToddlerBot is an open-source miniature humanoid (0.56 m, 3.4 kg) designed as a research platform that both runs policies and collects the data to train them, in simulation and in the real world. It is entirely 3D-printed around off-the-shelf Dynamixel motors and a Jetson Orin NX, for a total cost under 6,000 USD, 90% of which is motors and compute.
How it works
- Hardware. 30 active DoFs: 7 per arm, 6 per leg, 2 in the neck and 2 in the waist, packed in with spur gears, coupled bevel gears and parallel linkages. Two swappable end effectors (parallel-jaw gripper, compliant palm), two fisheye cameras, an IMU, a speaker and microphones.
- Digital twin. 3D-printed calibration devices set the motor zero points within a minute. A motor test bed and chirp-signal tracking data fit an actuation model per motor type; the parameters carry over to other motors of the same model.
- Sim data. Keyframe motions are tuned in a MuJoCo GUI tool and run open-loop; walking is a PPO policy rewarded for imitating a ZMP reference.
- Real data. A second upper body serves as leader arms, with a handheld gaming computer for walking, squatting, neck and waist. A two-layer PD controller keeps the lower body balanced. Diffusion policies are trained on the result.
Results
- Hugs objects up to 27 x 24 x 31 cm, lifts 1484 g (40% of its weight), and steps in place for up to 19 minutes without falling.
- Push-ups, pull-ups and RL walking transfer zero-shot; the fitted actuation model tracks with a 1.3° average error.
- With 60 demonstrations (collected in 20 minutes), diffusion policies succeed in 90% of 20 trials on bimanual and 75% on full-body manipulation, running onboard at about 100 ms latency.
- A student with no hardware experience built a second robot in three days; the manipulation policy reached the same 90% on it without new sysID.
Limitations
Off-the-shelf motors cap speed, torque and communication rate, which limits agile motions. The actuation model ignores motor temperature and is less accurate near the performance limit; heating caused more falls over time in the endurance test. The small scale restricts interaction with human-sized objects, and 3D-printed parts break more easily on impact than metal ones (up to 7 falls before breaking).
Videos
Figures
Abstract
Learning-based robotics research driven by data demands a new approach to robot hardware design-one that serves as both a platform for policy execution and a tool for embodied data collection to train policies. We introduce ToddlerBot, a low-cost, open-source humanoid robot platform designed for scalable policy learning and research in robotics and AI. ToddlerBot enables seamless acquisition of high-quality simulation and real-world data. The plug-and-play zero-point calibration and transferable motor system identification ensure a high-fidelity digital twin, enabling zero-shot policy transfer from simulation to the real world. A user-friendly teleoperation interface facilitates streamlined real-world data collection for learning motor skills from human demonstrations. Utilizing its data collection ability and anthropomorphic design, ToddlerBot is an ideal platform to perform whole-body loco-manipulation. Additionally, ToddlerBot's compact size (0.56m, 3.4kg) ensures safe operation in real-world environments. Reproducibility is achieved with an entirely 3D-printed, open-source design and commercially available components, keeping the total cost under 6,000 USD. Comprehensive documentation allows assembly and maintenance with basic technical expertise, as validated by a successful independent replication of the system. We demonstrate ToddlerBot's capabilities through arm span, payload, endurance tests, loco-manipulation tasks, and a collaborative long-horizon scenario where two robots tidy a toy session together. By advancing ML-compatibility, capability, and reproducibility, ToddlerBot provides a robust platform for scalable learning and dynamic policy execution in robotics research.
Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.