Galbot

Robotics · Humanoid Robots · Embodied AI

Launch

Galbot unveils ET1, claiming first self-learning humanoid robot

Its tennis debut aims to be humanoid robotics' AlphaGo moment, fusing split-second decision-making with fine motor control.

• Galbot unveiled the ET1, its first bipedal humanoid robot, at the World Robot Conference in Beijing on Aug 19, 2026, calling it the world's first autonomous self-learning humanoid robot.
• Galbot (also known as Galaxy General/Galaxy Universal) is a Beijing robotics startup that built its commercial reputation on wheeled, dual-arm platforms like the G1 used in retail and logistics before pivoting to a legged humanoid.
• ET1 runs AstraBrain-Agent, an embodied AI architecture that observes its surroundings, parses spoken instructions, and can learn new motor skills directly from a live human demonstration rather than pre-recorded training videos.
• The 173 cm, 65 kg prototype has 48 degrees of freedom (24 in the hands), runs on an NVIDIA Jetson Thor chip, and claims sub-100 ms glass-to-action latency, according to listed specs.
• At the reveal, ET1 performed a live breakdance routine, including a handstand mirrored in real time with a human dancer, and was slated to play tennis against athletes at the World Humanoid Robot Games opening on Aug 22.
• Galbot frames the tennis test as a potential 'AlphaGo moment' for humanoid robots, arguing that returning a serve requires both high-level strategy and precise, whole-body motor control in real time.
• The launch reflects a shift in the humanoid race from raw hardware specs toward embodied AI 'brains' that generalize skills without task-specific scripting, seen as the industry's real bottleneck.