Leeds, United Kingdom - July 10, 2025 A video shows a four-legged robot learning to walk across difficult terrain by adapting its gait in real time, a breakthrough developed by researchers from the University of Leeds and University College London (UCL). The robot, nicknamed “Clarence,” successfully crossed surfaces like muddy grass and piles of loose timber, even recovering from slips and trips. The animal-inspired AI system powering the robot enabled it to choose how to move depending on the terrain—without pre-programmed instructions. The research team created a framework that teaches the robot how to switch between gaits like trotting, bounding, and running, similar to how dogs, cats, or horses respond to unfamiliar environments. All training occurred in simulation. According to the researchers, Clarence learned the strategy in just nine hours and was later tested outdoors on terrain it had never seen before. First author Joseph Humphreys said: “Our findings could have a significant impact on the future of legged robot motion control by reducing many of the previous limitations around adaptability.” He added: “This deep reinforcement learning framework teaches gait strategies and behaviour inspired by real animals — or ‘bio-inspired’ — such as saving energy, adjusting movements as needed, and gait memory, to achieve highly adaptable and optimal movement, even in environments never previously encountered.” In testing, the robot navigated overgrown roots, woodchips, loose rocks, and timber. It was also deliberately knocked off balance with a sweeping brush to test recovery. Professor Zhou from UCL said: “Instead of training robots for specific tasks, we wanted to give them the strategic intelligence animals use to adapt their gaits — using principles like balance, coordination, and energy efficiency.” The system was developed without using sensors like sight or sound, and researchers say it opens new possibilities for robotics in hazardous environments, disaster zones, or even planetary exploration.