Galaxy General Shows One Brain Across Bipedal, Wheeled, and Heavy-Load Robots at WRC 2026

Overview

At the 2026 World Robot Conference (WRC), which opened on August 19, 2026 in Beijing Yizhuang, Galaxy General (Galbot) deployed three robot bodies on a single general-purpose brain, the AstraBrain-Agent:

• Galbot ET1 — bipedal, first public debut
• Galbot G1 — wheeled-arm robot, already in mass production
• Galbot S1 — heavy-load robot, running 7×24 hours in normal operation on CATL's production line

ET1 Bipedal Debut

After its Spring Festival Gala appearance, Galbot had mainly been known for the wheeled G1. ET1 is the company's first engineering closed loop combining bipedal locomotion with the same brain. The live demo was playing tennis—a task requiring whole-body coordination and real-time decision-making under adversarial conditions, far beyond pick-and-place.

• AstraBrain-Agent: handles task understanding, action planning, and native physical-world interaction
• Cerebellum module: motion control, translating high-level intent into joint trajectories
• Perceive-understand-decide-execute simultaneously: G1's capability signature in dynamic environments, already at industrial-grade commercial stability

ET1 is not competing with Optimus on showmanship. Its deliverable is that one brain can drive bodies of different forms—a feat few domestic companies have achieved across bipedal, wheeled, and heavy-load platforms simultaneously.

One Brain, Multiple Bodies

AstraBrain replaces the industry's default "one model per robot, one skill per scene" fragmentation. One general model simultaneously drives:

| Body Form | Commercial Scenario | Current Stage | |---|---|---| | Bipedal ET1 | Home service, cultural tourism, reception | WRC debut (8-19), skills in training (tennis, etc.) | | Wheeled-arm G1 | Industrial manufacturing, retail | Mature and mass-produced, used across customers | | Heavy-load S1 | 30 kg-class palletizing, logistics | 7×24 normal operation on CATL's production line |

While many international players have touted a "universal brain," this appears to be the first domestic case where one set of weights delivers productivity on three radically different bodies.

S1 at CATL: Production, Not a Demo

What S1 has already done:

• Depalletizing, carrying, and palletizing of 30 kg-class materials
• 7×24 normal operation on CATL's production line
• Partnerships with SAIC, BAIC, Bosch, Hyundai, and other industrial partners

The key detail for manufacturers: S1 at CATL is not a trial or a POC—it is normal operation. That means the line's process documents have already incorporated the robot's cycle time, failure modes, and downtime policies; otherwise CATL would not let it replace a fixed workstation worker.

What G1's Closed Loop Means for Engineers

G1 does not rely on pre-programmed responses to disturbances. It completes each operation autonomously via the same model's closed-loop capability in dynamic environments. This means:

• No need to pre-configure responses for every disturbance
• New objects and scenes require no retraining of dedicated models
• Deployers get a continuously adaptive general brain, not a toy that only knows one workstation

This is engineering proof of embodied intelligence moving toward physical-world generality: a general brain + general closed loop—the two capabilities that have most blocked humanoid robot commercialization for a decade.

Reasons for Caution

• ET1's tennis is only a technical demo; industrial-grade stability is still distant
• "One brain, many skills" has a sample size of three bodies; extending to robotic arms, quadrupeds, or cobots will be the real test over the next 12 months
• The 7×24 CATL figure lacks disclosed engineering metrics such as continuous runtime per station and mean time between failures

Still, the bottom line: a Chinese vendor has, for the first time, simultaneously claimed the three labels of general brain, cross-embodiment, and industrial normalization—an industrial milestone bigger than any PR video.

From "One Dish Per Robot" to One Chef for Every Cuisine

The next step for the industry is no longer "how many humanoids built" or "how fast they walk," but "how many robot forms one set of model weights can cover." If, within 12 months, AstraBrain-Agent can:

• Drive 5+ different body forms shipped by vendors,
• Win industrial-grade orders beyond heavy load (e.g., quadrupeds, collaborative arms),
• Be integrated by third-party robots without retraining,

then "general embodied intelligence" will stop being a slide-deck term and become infrastructure on the level of SageMaker or Kubernetes. The bell for that day has already rung—on CATL's production line.