The University of Michigan has developed Simulated humanoid robots that can learn to hike rugged terrain autonomously

According to foreign media reports, researchers at the University of Michigan have developed a new artificial intelligence (AI) model that can train humanoid robots for hiking, which is expected to accelerate the development of embodied intelligence and enable them to perform tasks such as autonomous search and rescue and ecological monitoring in undeveloped areas.

With the help of a new artificial intelligence framework called LEGO-H, researchers trained Unitree Robotics' humanoid robots equipped with cameras to plan routes in advance, avoid obstacles, maintain a balanced posture, and adjust speed and stride based on rugged terrain.

Traditionally, robots learn navigation on flat and accessible surfaces through pre built maps and continuous manual guidance, where high-level planning ("where to go") and low-level execution ("how to move") are considered independent problems.

Integrating navigation and movement into a single strategy learning framework enables robots to autonomously formulate movement strategies based on actual situations, without the need for pre programmed patterns by humans.

Compared with robots that have obtained perfect navigation and environmental information in advance, simulated autonomous robots are comparable or even superior in terms of efficiency and safety. Researchers say that its built-in body perception ability helps prevent injury, and removing this feature significantly reduces the success rate of hiking.

Virtual autonomous robots learn to adjust their body posture and movement based on terrain. For example, when encountering narrow spaces, robots learn to lean sideways to squeeze through obstacles. They can also choose a path based on obstacles, circling around high obstacles, crossing low obstacles, and if unable to cross, choosing to detour.

In this preliminary study, the upper body of the robot remained in a fixed state, as adding upper body movements would significantly increase modeling complexity. With the success of this concept validation study in leg movement, the research team is committed to achieving full body coordinated mountaineering to fully utilize the robot's full degrees of freedom, maximizing stability, safety, and efficiency during walking.