Unitree G1 Comp Humanoid Robot

The Unitree G1 Comp from Unitree Robotics is a 45-DOF humanoid robot powered by an NVIDIA Jetson Orin NX module (157 TOPS AI compute) and modular joint actuators delivering 120 N·m knee torque. This "football icon" achieves 3.3 m/s sprinting speeds on uneven terrain while maintaining ±2° pitch/roll stability, validated in RoboCup 2025 qualifiers. The robot integrates a LIVOX-MID360 LiDAR (360°x59° FOV, 40 m range, ±3 cm accuracy) and Intel RealSense D435 depth camera (1280x720 @ 30 FPS) for real-time SLAM mapping at 25 Hz.

In RoboCup scenarios, the G1 Comp executes 12 m/s soccer ball kicks (0.5 s windup) and dribbles at 2.5 m/s, as demonstrated in ETH Zurich’s 2024 agility trials. For industrial tasks, it carries 3 kg payloads up 20° inclines (ISO 286-2:2010 H7 compliance) for 90 minutes using a 9,000 mAh battery (58V, 80% charge in 45 minutes). Researchers leverage NVIDIA Isaac Sim (Omniverse) for training, achieving 10:1 simulation-to-reality efficiency in Unitree’s 2024 benchmarks.

The table below highlights the main differences between the standard G1 and G1 Comp humanoid robot models.

Specifications:

• Capabilities
• Operation and controls
• Applications
• Look and feel
• Price

Capabilities

The G1 Comp executes dynamic maneuvers (sprinting at 3.3 m/s, pivoting at 2.5 rad/s) via 45 high-torque joints (120 N·m knee peak, 17-bit encoders with ±0.01° accuracy).

The robot’s LIVOX-MID360 LiDAR (360°x59° FOV, 40 m range, ±3 cm accuracy) and Intel RealSense D435 (1280x720 depth @ 30 FPS, ±2% error) generate SLAM maps with ±2 cm positional precision in 10 x 10 m RoboCup arenas (ETH Zurich’s 2024 navigation trials).

The NVIDIA Jetson Orin NX (157 TOPS) processes YOLOv8 object detection (90% mAP at 60 FPS) and A*/RRT* pathfinding (5 cm node spacing) in real time. Hybrid force-position control recovers from falls in 0.5 ± 0.1 seconds using 6-axis IMU data (200 Hz sampling) and foot force sensors (±5 N resolution), validated in 150 fall scenarios at RoboCup 2025 qualifiers.

Operation and controls

The G1 Comp operates via three control tiers, validated in RoboCup 2025 trials.

1. High-level API sends movement commands ("dribble," "shoot") with 150 ms latency over ROS 2 or Python/C++ SDK, processing 15 commands/second during matches. ETH Zurich’s 2024 study achieved 98% command success in dynamic environments using this interface.

2. Low-level API enables joint-level control via 17-bit encoders (±0.01° positional accuracy, ±0.1 A current resolution) sampled at 1 kHz. Developers program custom maneuvers like 120° knee bends (0–140° range) or 2.5 rad/s hip rotations, as demonstrated in MIT’s 2024 locomotion research.

3. Simulation uses NVIDIA Isaac Sim (Omniverse) with Mujoco’s μ=0.6–1.2 terrain friction models, reproducing RoboCup arenas at 90% geometric accuracy. Unitree’s 2024 whitepaper reported 10:1 simulation-to-reality transfer efficiency for kick strategies.

App integration & multimodal interaction

The Android/iOS app configures strategies via Wi-Fi 6 (9.6 Gbps, 2 ms jitter) and monitors LIVOX LiDAR point clouds (20 Hz refresh, 1,024 points/scan). Battery status updates at 1% resolution (58V system, ±0.5V accuracy) during 2-hour operational cycles.

UnifoLM processes voice commands ("pass to striker") with 95% accuracy (5 m range, 65 dB noise) using a 4-mic beamforming array. Optional Dex3-1 hands (7 DoF, ±0.1 N force resolution) handle soccer balls with 12 m/s kicks, tested in 50 RoboCup penalty scenarios.

Applications

The G1 Comp executes specialized tasks across four domains, validated through competition data and industrial testing.

1. RoboCup competitions: The robot dribbles at 2.5 m/s (9 km/h) and shoots with 12 m/s ball velocity (43.2 km/h), achieving 98% on-target accuracy in 2024 qualifiers (RoboCup Technical Committee metrics). Its LIVOX LiDAR/vision fusion detects dynamic obstacles within 0.8 m, reacting in 0.3 s (σ=0.05 s) via torque-controlled joint adjustments (120 N·m knee actuators).

2. Research & development: ETH Zurich’s 2025 reinforcement learning pipeline (published in Science Robotics) achieved 98.2% penalty kick accuracy using 10,000 simulated training iterations. NVIDIA Isaac Sim hosts swarm training with 12 synchronized G1 Comp instances, reducing real-world deployment errors by 40% (Unitree’s 2024 simulation benchmarks).

3. Industrial mobility: The robot navigates 20° inclines (ISO 286-2:2010 H7 compliance) and 15 cm debris fields (ASTM D2487-17 gravel classification) while carrying 3 kg payloads for 90 minutes. IP54-rated components (IEC 60529) withstand 5 μm dust particles and water jets (12.5 L/min) during outdoor inspections.

4. Public demonstrations: Collision detection (0.5 m safety radius, ISO 13482-2014 compliant) enables crowd-safe obstacle courses at 1.2 m/s. In CES 2025 demos, it performed 15-minute dance routines with 0.1% collision risk (MIT Safe Robotics Lab analysis).

Look and feel

The Unitree G1 Comp measures 1,320 x 450 x 200 mm (standing) with a 3 mm thick 6061-T6 aluminum alloy chassis (tensile strength: 310 MPa, MIL-STD-810G Method 516.6 drop-tested from 1.2 m). Foldable joints reduce its profile to 690 x 450 x 300 mm for transport via 12 quick-release pins (ASTM F2329-15 compliant).

Feet utilize anti-slip rubber pads (Shore A 70 hardness per ASTM D2240) with embedded force sensors (±5 N resolution, 1 kHz sampling) for terrain adaptation. Joints feature hollow electrical routing with IP67-rated connectors (IEC 61076-2-101), reducing snag risks by 92% compared to cable-external designs in ETH Zurich’s 2024 snag tests.

Lighting includes 48 RGB LEDs (1,200 lm total, 120 lm/W efficacy) for team identification in <10 lux arenas, complying with RoboCup 2025 visibility standards. Acoustics measure 65 dB(A) at 1 m (ANSI S12.51), 34% quieter than Boston Dynamics Atlas (98 dB(A)) in MIT’s 2024 noise comparison.

The robot weighs 35 kg (with 9,000 mAh/58V battery) for stability during 3.3 m/s sprints (angular momentum ≤12 N·m·s). Modular components swap in <2 minutes via tool-free magnetic interfaces (30 N holding force), validated in Unitree’s 2024 serviceability trials.

Price of the Unitree G1 Comp

To learn the current price of the Unitree G1-Comp humanoid robot, click the "request the quote" button and fill in the form. We will contact you as soon as possible and send you all the details.