Booster T1

The Booster T1 is an advanced, open-source humanoid robot platform engineered by Booster Robotics, specifically tailored for developers, researchers, and competitive robotics teams. Standing at 118 cm tall and weighing approximately 30 kg, it features a modular bipedal architecture with 23 degrees of freedom (DoF) in its base configuration, expandable to 31 with grippers or 41 with dexterous hands. The design emphasizes lightweight construction, flexibility, and exceptional durability, enabling complex dynamic movements such as omnidirectional walking, sports maneuvers (e.g., soccer kicks), intricate kung fu sequences, and robust fall recovery. Its skeleton comprises a head (2 DoF: yaw and pitch), torso with waist yaw (1 DoF), dual arms (4 DoF each: shoulder pitch/roll/yaw, elbow; extensible), and legs (6 DoF each: hip pitch/roll/yaw, knee, ankle pitch/roll). Joints offer extensive ranges, e.g., hip pitch ±118°, knee 0-123°, with custom Booster motors delivering peak torque up to 130 N·m, dual encoders for precise position/velocity feedback, and hollow-axis design for compactness. At the core of its computational architecture lies a dual-board system: a Motion Control Board powered by Intel Core i7 (Express-i7 with 6P+8E cores up to 4.8 GHz, 8GB RAM, 512GB storage) handling real-time kinematics and dynamics, paired with a Perception Board featuring NVIDIA Jetson AGX Orin 32GB (8-core ARM Cortex-A78AE @2GHz, 200 TOPS AI via Tensor Cores, 32GB RAM, 512GB storage). This setup supports high-fidelity AI inference, reinforcement learning (RL), and multimodal processing. The software stack is ROS2-compatible via Booster Robotics SDK (GitHub: BoosterRobotics/booster_robotics_sdk), with Fast-DDS for low-latency pub/sub and RPC interfaces. Key topics include /tf transforms, joint control (/rt/joint_ctrl), low-state feedback (/rt/low_state with IMU/motor data), and vision streams. Open-source tools like Booster GYM enable RL-based locomotion training, compatible with Isaac Sim and MuJoCo simulations. Sensor suite includes Intel RealSense D455 or Orbbec Gemini 2 355L depth cameras (RGB-D, global shutter IR, 6-axis IMU integration), providing rich point clouds for perception, SLAM, and object detection; a 9-axis IMU for balance/orientation; circular 6-mic array and speaker for audio interaction (e.g., real-time voice with Doubao LLM); optional 3D LiDAR for 360° FOV mapping. Connectivity via WiFi 6, Bluetooth 5.2, dual Gigabit Ethernet (IP: 192.168.10.x), USB/Ethernet expansions, and mobile app/joystick control. Battery is a 10.5 Ah Li-ion pack (torso-mounted, ≤2h charge, ≥500 cycles), yielding 1.5-2 hours walking (0.5 m/s max) or 4 hours standing autonomy, with low-battery/fall alerts and thermal overheat protection. Real-world deployments underscore its prowess: Champion of RoboCup 2025 AdultSize Soccer (Hephaestus Team, Tsinghua University), sweeping medals at World Humanoid Robotics Games Federation 2025; adopted by 50+ teams/institutes including Purdue (table tennis), CMU (FALCON RL), UC Berkeley (FastTD3 RL), Frodobots (fighting games). NVIDIA-endorsed (signed by Jensen Huang), it's now shipping globally for research, competitions, and embodied AI development, bridging simulation-to-real gaps with robust hardware-software integration.