• Agibot G2 Wheeled Humanoid Robot
• 26 DoF base body, expandable to ~50 DoF with optional dexterous hands
• Dual 7-DoF arms with full-joint torque sensing and sub-millimeter precision
• NVIDIA Jetson Thor T5000 computing board delivering up to 2070 TFLOPS
• Dual hot-swappable batteries with 4h+ runtime and autonomous recharging
• Omnidirectional wheeled chassis with in-place rotation and 1.5 m/s travel speed
• Optional dexterous hands with up to 12 active DoF and 3D tactile sensing

The Agibot G2 Wheeled Humanoid Robot is an industrial-grade embodied intelligence platform developed by AgiBot. Designed for precision manipulation and complex task execution, the G2 combines a full-humanoid upper body with an omnidirectional wheeled base, enabling flexible deployment across structured indoor environments. Its dual-arm configuration achieves sub-millimeter assembly accuracy through full-joint torque control, making it suitable for applications that demand both repeatability and dexterity. Built with automotive-grade components and IP42 protection, the platform is engineered for sustained real-world operation rather than laboratory demonstration.

The G2 processes its environment using a multi-modal sensor array that includes RGB-D cameras, wrist-mounted depth sensors, dual LiDAR units with 360-degree coverage, a fisheye camera array, and 8 ultrasonic sensors for close-range obstacle detection. Interaction is handled through a tri-modal expression system combining voice conversation, gesture recognition, and an animated facial display capable of conveying emotional cues and supports interactive multi-user engagement. Deployment flexibility is supported through the Genie RL reinforcement learning toolchain, which is designed to support efficient task training, and an LLM-integrated knowledge base that allows real-time role switching. The G2 is designed to serve commercial, industrial, and research organizations seeking a production-ready autonomous robot platform. AI-driven capabilities depend on deployed software configuration and use-case implementation.