Zeroth-Bot is an open-source, 3D-printed humanoid robot platform designed for research and development in robotics, particularly focusing on sim-to-real transfer and reinforcement learning (RL). It aims to provide an accessible and affordable platform for experimenting with advanced robotic control algorithms. The project provides hardware, software, and simulation environments to accelerate robotic AI development.
Zeroth-Bot distinguishes itself through its complete open-source nature, encompassing hardware designs, software SDK, and simulation environments. It provides a low-cost entry point into humanoid robotics with a starting BOM of $350. The project's focus on a streamlined sim-to-real pipeline and integrated RL training tools further sets it apart.
- Hardware Drivers: Supports servo control, IMU interfaces, and actuator management for flexible robot operation.
- Python API: Provides an intuitive Python API (
kos_zbotpackage) for simplified development and control. - RL Training: Facilitates GPU-accelerated RL for whole-body control, enabling advanced robotic behaviors.
- Sim-to-Real: Includes SysID-calibrated actuators and pre-loaded robot models for seamless transfer from simulation to physical robot.
- Real-time Control: Enables low-latency servo communication and sensor data processing for responsive control.
- CLI Tools: Offers command-line utilities for robot calibration, diagnostics, and system configuration.
- Open-Source: Hardware, software, and simulation tools are all available under an open-source license for community contribution.
Zeroth-Bot is currently in public beta, offering basic capabilities like locomotion, vision, and speech. The project is actively developed and maintained by the open-source community, with regular updates and ongoing improvements. While still considered work in progress, the project aims for a stable V1.0 release in June 2025, indicating a commitment to long-term support and reliability.
Zeroth-Bot benefits roboticists, AI researchers, and developers seeking a readily available platform for experimenting with humanoid robotics and reinforcement learning. It addresses the challenges of sim-to-real transfer and provides a cost-effective alternative to proprietary robotic platforms, empowering broader access to advanced robotic research and development.