Scientists from the University of Hong Kong have created an advanced terrain-mapping model that enables four-legged robots to navigate highly complex environments without human guidance. Their robot, trained in simulation and tested on a Unitree Go1 platform, uses lidar data to build a multilayer elevation map of its surroundings. This map captures the height, depth, and structure of obstacles, allowing the robot to understand and move through challenging spaces.
The robot can autonomously switch between multiple movement skills such as crawling under overhangs, jumping across gaps, and climbing over raised surfaces depending on what the terrain demands. Even more impressively, although it was never explicitly programmed for path planning, it naturally finds alternative routes when it encounters an obstacle too tall or too wide, showing a form of emergent problem-solving.
While the system’s strength lies in its ability to handle diverse terrain, it remains limited to the data it was trained on and cannot yet learn directly from real-world experience. The researchers plan to improve this by integrating real-world data and hope to commercialize the system for inspection tasks in environments like construction sites.
Read more-https://spectrum.ieee.org/robot-navigation-3d-mapping