• 2 min read
KAIST robot switches gaits on the fly at 6 m/s
KAIST’s four-legged HOUND uses a single controller to walk, run, jump, and clear obstacles in real outdoor terrain at up to 6 m/s.
Image: TechXplore
A KAIST research team says it has built a four-legged robot that can judge terrain in real time and switch between walking, running, jumping, and other movement skills with a single controller. The work, led by Professor Hae-Won Park of the Korea Advanced Institute of Science and Technology (KAIST) Department of Mechanical Engineering, was published in Science Robotics.
Unlike earlier quadruped systems that handled each gait separately, the new control approach is designed to let the robot change locomotion strategy as conditions shift, from stairs and ledges to gaps, stepping stones, and forest trails. The team says that has been a major limitation for legged robots operating outside controlled settings.
The system, called APT-RL—short for Action Pretrained Transformer-based Reinforcement Learning—first learns a range of locomotion skills and then combines and transitions between them as needed in the real world.
Rather than relying on motion capture from people or animals, the researchers generated 15.5 hours of gait training data entirely in computer simulation in eight minutes. They used that data to teach basic movement capabilities using robot dynamics and trajectory optimization, then applied reinforcement learning so the machine could choose the best gait for complex three-dimensional terrain.
To perceive its surroundings, the robot combines a depth camera with LiDAR, allowing it to detect the environment and target speed in real time and pick an appropriate locomotion strategy.
Recommended reading
Satellite startups pull $8.1B in half a year
Outdoor tests on campus and forest trails
The team tested its robot, KAIST HOUND, on indoor obstacle courses and in outdoor settings including KAIST’s campus and forest trails. According to the researchers, it stayed stable on urban terrain such as stairs, grass, and slopes, and on natural terrain with fallen trees, exposed roots, and leaf-covered paths.
In rugged terrain, the robot reached a peak instantaneous speed of six meters per second—about 22 kilometers per hour. The experiments showed it could autonomously switch between a trot and a bound, while integrating walking, running, jumping, and ledge-clearing under one controller.
“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.”
Jun-Gill Kang, who was affiliated with the Agency for Defense Development (ADD) at the time of the research, and Jaehyun Park, a Ph.D. candidate in KAIST’s Department of Mechanical Engineering, are co-first authors. Professor Seungwoo Hong of Korea University and Park are co-corresponding authors.
The paper is “Agile perceptive multiskill locomotion for quadrupedal robots in the wild,” by Jun-Gill Kang et al, published in Science Robotics (2026). DOI: 10.1126/scirobotics.adz7397.
Frontier Editor
Dan is our resident futurist, covering electric mobility, space exploration, and the smart home. He's interested in atoms just as much as bits. Whether it's a new battery chemistry, a reusable rocket, or a protocol that finally makes IoT devices talk to each other, Dan breaks down the engineering that pushes humanity forward.
via TechXplore
