Is the frontier of humanoid robotics shifting from hardware to intelligence?

Galbot’s ET1 humanoid robot, showcased at the World Robot Conference, highlights a critical shift in Asian robotics from hardware prowess to embodied intelligence. While Chinese firms like Unitree Robotics have seen valuations soar on physical performance, Galbot's focus on its AstraBrain foundation model suggests that what a robot can understand and learn is becoming more important than how fast it can move. This approach, centered on zero-shot generalization and user-friendly teaching, aims to make robots adaptable to varied tasks without extensive retraining, a key hurdle for broader adoption in Asia's diverse industrial and retail settings. The company’s strategy to open its platforms and models to partners is a calculated move to accelerate data collection. Real-world deployments in smart pharmacies and retail generate complex data that simulations alone cannot replicate. This data flywheel, where deployments improve models and better models ease new deployments, could give Galbot a significant edge in developing robust, general-purpose AI for physical environments. The challenge lies in scaling this data collection and ensuring the reliability needed for commercial use, where single demonstrations are insufficient. The thing to watch for Chinese robotics is whether Galbot can truly achieve its benchmark of 70-80% success rates for zero-shot generalization and low-cost post-training by ordinary users. If successful, this could democratize robot deployment, making advanced automation accessible to smaller businesses across Asia and potentially driving a new wave of industrial transformation beyond current specialized applications.