KAIST unveils autonomous quadruped robot that adapts locomotion to terrain

Researchers at South Korea's Korea Advanced Institute of Science and Technology (KAIST) have developed a four‑legged robot that can analyse its surroundings and switch in real time between walking, running and jumping. The control system, called APT‑RL (Action Pretrained Transformer‑based Reinforcement Learning), learns a repertoire of locomotion skills in simulation and then combines them autonomously based on terrain cues such as stairs, stones, gaps or forest paths.

In outdoor tests the robot, named KAIST Hound, demonstrated stable movement at speeds up to 22 km/h (6 m/s), selecting appropriate gait patterns like trot or bound without human intervention. The work was published in the July issue of Science Robotics and is presented as a step toward physical‑AI robots for disaster response, defence missions and industrial inspection in rugged environments.