Researchers at ArcLab at the University of Hong Kong have developed a robops that can walk on two legs over difficult terrain, maintain balance during impacts, and even get up on its own after falling, reports Interesting Engineering.
The uniqueness of this development lies in the use of a bio-inspired controller TumblerNet, which works on the basis of Deep Reinforcement Learning. It allows the robot to imitate human gait and adapt to different conditions - even on sand, loose stones or mats.
TumblerNet combines information about the center of mass and center of pressure into a single system that mimics human balancing. This allows the robot to freely switch between quadrupedal and bipedal modes, using its front or hind legs, and to perform complex movements, including U-turns and circular movements.
Another feature is its durability . It can withstand impacts, pushes, and even falls — and it doesn’t need an additional balancing system. If it loses its balance, the robot can pick itself up on its own, even if it gets caught on objects.
The researchers also tested the robot on a beach, an environment where it is difficult to maintain balance. Despite the challenging conditions, it successfully moved on two legs, demonstrating the high adaptability and reliability of the development.
Unlike traditional four-legged dog robots, this model is better adapted to human environments. It can climb stairs, avoid obstacles, and leave its "paws free" to perform tasks. This makes it particularly useful for human care, rescue, and human interaction in real-world environments.