A time is rapidly approaching when robots can climb walls, leap across stepping stones, and navigate complex environments freely like humans. Domestic researchers have developed a technology that allows quadruped robots to reliably find safe foot placement and accurately land at desired points even while moving quickly using artificial intelligence (AI).
Professor Hwang Bo-je-min and his research team at the Korea Advanced Institute of Science and Technology (KAIST) noted on the 29th that they developed a new control system enabling quadruped robots to swiftly traverse challenging terrains akin to parkour. Parkour refers to efficiently moving using physical structures or objects such as buildings, bridges, and walls.
The research team designed a control strategy with two modules called 'Planner' and 'Tracker'. The Planner utilizes AI to calculate several candidate paths that the robot can safely step on, filtering out dangerous or impossible routes in advance. It then simulates the remaining candidate paths to select a safe and efficient route.
The Tracker guides the robot to step on the target point determined by the Planner. The research team also used algorithms that automatically generated increasingly complex environments to train the Tracker. As a result, the robot was able to learn challenging movements.
Professor Hwang Bo-je-min said, "Robots are just like humans. They should learn gradually by increasing the difficulty of tasks rather than being given difficult tasks from the start," and added, "All calculations necessary for the Planner and Tracker were performed in real time on the computer onboard the robot."
The research team conducted experiments indoors using their self-developed quadruped robot 'Raibo'. Raibo, equipped with the new control system, ran while stepping against walls, jumped over stepping stones at a speed of 4 meters per second, and cleared wide gaps of up to 1.3 meters. It moved swiftly and accurately on terrains made of slopes, stairs, and boxes of various sizes. It showed a high success rate even when feet slipped, stepping platforms wobbled, or when there were errors in the map information.
It is expected that the system developed this time will allow robots to maneuver flexibly in unpredictable terrains, such as actual disaster sites or construction sites. The research team stated, "This study has laid a technological foundation for robots to work in hazardous areas instead of humans," and noted, "We plan to enhance this system for application in outdoor environments to significantly increase its utility in rescue operations, exploration, and military fields."
The results of this research were published today in the international academic journal 'Science Robotics'.
References
Science robotics (2025), DOI: https://doi.org/10.1126/scirobotics.ads6192