KAIST researchers developed a four-legged robot that can independently choose whether to walk, run, or jump as terrain changes, eliminating the need to switch between separate control programs. The system, called Action Pretrained Transformer-based Reinforcement Learning (APT-RL), was trained largely through simulation and uses depth cameras and LiDAR to navigate stairs, slopes, gaps, forest trails, and other difficult environments. Tested on the KAIST HOUND robot, the technology reached speeds of up to 6 meters per second, or about 22 kilometers per hour, and could eventually support search-and-rescue missions, industrial inspections, defense operations, and other work in hazardous terrain. The advance, led by Professor Hae-Won Park of the Department of Mechanical Engineering, marks a shift in how robots interact with real environments. Instead of following fixed movement rules, the robot adapts in real time, choosing the most effective way to move. “We expect this to become a foundational technology that expands the potential uses of physical-AI-based walking robots in rugged environments such as disaster sites, defense missions, and industrial facility inspections,” Park said. The researchers developed a learning-based control system that allows the robot to learn multiple movement skills and combine them as needed, blending walking, running, and jumping into a unified framework.