Maker H01

Maker H01, developed by GigaAI (极佳视界), a leading Chinese embodied intelligence startup backed by Huawei Hubble Investment and others, is a wheeled humanoid robot unveiled on November 26, 2025. Positioned as a 'Physical AGI Native Platform,' it bridges advanced AI models with physical embodiment, facilitating 'model-to-body' convergence for real-world tasks. Departing from bipedal designs, its omnidirectional all-wheel-drive wheeled chassis prioritizes stability, speed (up to 1.5 m/s or ~5.4 km/h, with reports up to 8 km/h), and endurance (4 hours per charge), ideal for dynamic environments like factories, homes, and services. The architecture integrates dual 7-DOF bionic arms (total 28 DOF excluding end-effectors) with a vertical reach of 0-2 meters, supporting 5 kg maximum payload per arm (3 kg rated). Modular end-effectors, such as multi-finger grippers or dexterous hands, enable precise manipulation for tasks including pick-and-place, stacking, pouring, and collaborative operations. The upper body emphasizes 'operation and upper limbs' for superior physical interaction and data collection. Perception relies on a sophisticated multi-modal sensor array: 5 RGB cameras and 4 RGBD cameras across head, chest, and hands for visual-depth fusion, plus chassis-mounted 360° LiDAR and ultrasonic sensors for omnidirectional obstacle avoidance and safe human interaction (IP20 protection). This rich sensing feeds into low-latency (250-450 ms glass-to-action) control. Central to its intelligence is GigaAI's full-stack software ecosystem. GigaBrain-0, an open-source Vision-Language-Action (VLA) model, acts as the 'embodied brain,' delivering end-to-end decisions from multi-modal inputs (images, depth, text, states) to structured task/motion planning. It excels in 3D spatial reasoning, long-horizon tasks (e.g., desk tidying, coffee making), and generalization, outperforming SOTA benchmarks. GigaWorld-0, the world model platform, generates high-fidelity synthetic data (~300% generalization boost), resolving data costs, Sim2Real gaps, and simulator errors via physics-accurate simulations. Their synergy with reinforcement learning targets 95% success across 90% of common tasks, embodying the 'World Model + Action Model + RL' paradigm. Hardware specs include ~162 cm height, 88 kg weight (aluminum alloy frames + ABS shells), Linux-based OS with custom stacks, and connectivity (Bluetooth, Ethernet, Wi-Fi). No specific CPU/GPU or battery kWh/voltage disclosed. Deployment is accelerating: prototype stage with mass production underway, partnerships with top OEMs, humanoid centers, and scenario leaders in industry/service/home. Funded by 500M+ CNY in 3 months, it targets home service, hospitality, logistics, education, and R&D, heralding physical AI's 'ChatGPT moment' in 2-3 years.