ToddlerBot: Open-Source ML-Compatible Humanoid Platform for Loco-Manipulation

TL;DR

ToddlerBot addresses the data-hungry nature of learning-based robotics by delivering an open-source, ML-compatible humanoid platform that integrates scalable simulation data with real-world data collection. Its high-fidelity digital twin, zero-point calibration, and motor sysID enable zero-shot sim-to-real transfer for whole-body loco-manipulation, all in a compact, low-cost package. The paper demonstrates broad capabilities through diverse tasks and confirms reproducibility via independent replications and open assembly instructions, making ML-driven robotics research more accessible. Collectively, ToddlerBot provides a practical, extensible platform for scalable policy learning, robust data collection, and community-driven advancement in humanoid robotics.

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.

Figures (16)

• Figure 1: ToddlerBot is an open-source humanoid platform for large-scale, high-quality data collection in both simulation and the real world. It combines massive parallel simulation and an accurate digital twin for simulation, and an intuitive teleoperation device for whole-body control in the real world. ToddlerBot enables diverse loco-manipulation skills, including walking, push-ups, pull-ups, wagon pushing, bimanual, and full-body manipulation, learned from both data sources.
• Figure 2: Mechatronic Design. Orange markers highlights ToddlerBot's 30 active DoFs: 7 per arm, 6 per leg, a 2 on neck, and a 2 on waist. Green markers indicate two end-effector designs—a compliant palm and a parallel-jaw gripper. Purple markers denote the electronics layout with exploded views, featuring 2 fisheye cameras, 1 speaker, 2 microphones, 1 IMU, and 1 Jetson Orin computer.
• Figure 3: Arm Span, Payload, and Trajectory Tracking. On the left, we show that with a torso dimension of $13 \times 9 \times 12~\mathrm{cm}^3$, ToddlerBot can grasp objects up to $27 \times 24 \times 31~\mathrm{cm}^3$, about 14 times the torso size. Additionally, ToddlerBot can lift weights up to $1484~\mathrm{g}$, which is 40% of its body weight ($3484~\mathrm{g}$). On the right, we present ten consecutive real-world rollouts of an RL walking policy tracking a square trajectory with a predefined velocity profile. Both raw and smoothed linear and angular velocity tracking are displayed, with real-world results averaged across trials.
• Figure 4: Experiment Results. We present four different tasks: push-up, pull-up, bimanual, and full-body manipulation, showing ToddlerBot's capability in challenging loco-manipulation tasks.
• Figure 5: Long-horizon Collaboration. In this task, two instances of ToddlerBot, Arya and Toddy, collaborate to clean up a toy session. (1) The task begins with a pink octopus on the table and a purple octopus on the ground. (2) Arya picks up the pink octopus from the table and places it in the wagon. (3) Arya walks to the wagon handle. (4) Arya grasps the handle while Toddy walks over. (5) Arya pushes the wagon toward the purple octopus. (6) Toddy reaches the pickup position. (7) Toddy kneels and picks up the purple octopus. (8) Finally, Arya and Toddy leave side by side.