Chinese facility uses trial and error to train machines
Researchers at Tashan built a specialized environment where robotic models develop physical skills through repeated practice rather than rigid programming. The long term objective is enabling trained systems to share learned neural representations with new devices instantly.
The Blend
A Chinese tech company called Tashan Technology has built a specialized testing space where robots learn physical tasks through active trial and error instead of pre-written software instructions. Rather than following rigid code, these experimental models repeatedly practice movements, fail, and adjust their actions based on what went wrong.
Most modern automated systems struggle when faced with situations their programmers did not explicitly anticipate. By allowing machines to analyze failures, such as hitting an obstacle and calculating a new path, engineers hope to create more resilient hardware. As researcher Wang Peng noted to Azernews, this hands-on exploration could help systems develop a more versatile understanding of how physical objects interact.
Looking ahead, researchers want these systems to upload their learned skills to a central network. If one unit figures out how to navigate a tricky physical space, other devices could instantly adopt that knowledge without having to repeat the same practice sessions.
While this training method mirrors how human children learn about the world, physical wear and tear remains a significant hurdle. It is still unclear whether delicate mechanical components can endure thousands of collisions during the learning phase without triggering unsustainable maintenance costs.
Written independently by AI News Smoothie from the reporting listed below. Facts belong to the original publishers. Follow the links for their full coverage.
Ingredients
- China opens “kindergarten” for robots
Tashan Technology has built an experimental facility where robots learn physical tasks by practicing and making mistakes instead of following strict pre-programmed code.