Galbot ET1 Debuts at WRC 2026 Ahead of Human–Robot Tennis Showcase

Galbot introduced Galbot ET1, its first bipedal humanoid robot, at the 2026 World Robot Conference in Beijing. The company presented ET1 alongside its AstraBrain embodied-AI system as a robot built to perceive, interact and learn in the physical world—not simply replay fixed motion routines.

The robot’s next public test was scheduled for the opening of the World Humanoid Robot Games on Aug. 22. Galbot said ET1 would take part in a human–robot mixed-doubles tennis performance and interact with tennis guests during the opening event.

Why tennis is such a difficult humanoid-robot test

A tennis rally compresses several hard problems into a single, fast-moving task. A humanoid must move into position while maintaining whole-body balance, track the ball and opponent, control the racket and wrist precisely, and decide where and how to return the shot.

Galbot’s demonstrations and explanations emphasize several requirements:

• Dynamic balance: The robot must run, stop and change direction without losing stability.
• Fine motor control: The racket needs accurate wrist and arm positioning at the moment of contact.
• Real-time perception: The robot must continuously interpret the ball, opponent and changing court conditions.
• Strategy: It must choose actions such as sending the ball left or right, or placing it short or deep, rather than repeating one predetermined swing.

Galbot describes the underlying system through a division between a higher-level “cerebrum,” which handles perception, reasoning and planning, and a “cerebellum,” which converts those intentions into continuous, stable full-body movement.

That combination matters because a robot can look athletic in a scripted demonstration without being able to adapt when the ball, opponent or timing changes. Tennis is therefore a useful stress test for whether perception, decision-making and movement control operate as one closed loop.

Could autonomous tennis create an “AlphaGo moment”?

Galbot and commentators have framed autonomous humanoid tennis as a possible “AlphaGo moment” for motion intelligence: a visible milestone in which a machine does more than execute movements and instead selects actions while moving through an uncertain environment.

The comparison should be treated carefully. A successful match or rally would show an impressive combination of perception, tactics and physical control, but it would not by itself establish general-purpose intelligence. The strongest conclusion supported by the available reporting is narrower: robust, unscripted tennis would be evidence of progress toward humanoid robots that can adaptively “think while moving.”

How AstraBrain-Agent is intended to work

ET1 is equipped with Galbot’s self-developed AstraBrain-Agent, described as a physical-world agent for real-time multimodal interaction and motion control. According to the company and event reporting, it can continuously observe its surroundings, understand spoken or natural-language instructions, and dynamically plan behavior and movement sequences.

The system is intended to connect three capabilities:

1. See: Gather information from the surrounding physical environment.
2. Understand and plan: Interpret a person’s instruction and determine an appropriate course of action.
3. Act and learn: Turn that plan into coordinated movement and acquire new skills through interaction.

Galbot said ET1 can learn new movement skills without relying on motion-capture data or prerecorded training videos. The company also demonstrated real-time interaction in which the robot made eye contact, moved its head and generated body movements in response to language.

This is the distinction Galbot is trying to draw between a conventional scripted robot and an embodied agent: the latter is expected to respond to what is happening now, revise its plan and build on interaction. Those capabilities remain company-reported demonstrations, so their performance outside controlled settings is still an open question.

The household work shown at WRC 2026

The household demonstrations at the event were performed by Galbot’s G1 robots, rather than being described as ET1’s complete domestic product portfolio. They included putting objects away, washing and hanging laundry, and folding clothes on a bed. Other reporting also described breakfast-preparation tasks such as making toast, pouring drinks and coordinating both arms.

These tasks are more technically demanding than they appear. They involve recognizing different objects, handling items that may roll, swing or deform, manipulating flexible clothing and bedding, and completing long sequences in the correct order. The demonstrations were presented as examples of AstraBrain’s broader goal: using a shared AI system across robots with different bodies and applications, including homes, retail and industrial environments.

What the ET1 unveiling means

ET1’s importance is less about its ability to swing a tennis racket than about the architecture Galbot is presenting behind it. The company wants one embodied-AI system to connect language, perception, planning and whole-body control across multiple robot forms.

The Aug. 22 mixed-doubles appearance was designed to make that ambition visible. If ET1 can respond reliably to changing play and human partners, it will offer a compelling demonstration of adaptive physical intelligence. But the lasting test will be whether the same closed-loop abilities transfer from a staged tennis showcase to the unpredictable, repetitive tasks of everyday work.