Next-generation humanoid robot developed for extreme environments

KAIST develops an AI-enabled bipedal robot capable of running, climbing steps and executing tasks in industrial environments and lunar simulations.
KAIST presenta un robot humanoide de próxima generación con IA

Capable of walking on uneven terrain and with the potential to operate in complex lunar and industrial environments, KAIST introduces a new humanoid robot that marks a breakthrough in robotic autonomy and control.

Full control from hardware to artificial intelligence

This system stands out for its mobility performance, with a maximum speed of 12 km/h and the ability to overcome steps of up to 30 cm.

The robot’s design, with human proportions (165 cm in height and 75 kg in weight), is intended to operate in human-centered spaces. KAIST has manufactured all the essential components, from motors to controllers, consolidating its technological independence in the field of robotic hardware.

From simulation to reality: a humanoid robot

The implementation of an artificial intelligence controller trained by reinforcement learning in simulated environments has been one of the most relevant aspects, achieving later its successful application in real scenarios.

The humanoid robot has been tested with complex movements, demonstrating its dynamic stability even without visual feedback. In addition, it can run continuously and maintain balance on uneven surfaces.

International cooperation and development of complete systems

This breakthrough is part of a larger project that aims to build a complete humanoid robot. To do so, KAIST is collaborating with MIT and other internal divisions such as the Kim Jaechul School of AI and the Department of Electrical Engineering. The goal is to integrate arms, hands, navigation and manipulation to create a system capable of operating doors, carrying heavy objects or manipulating valves while walking.

Acrobatic maneuvers of Jumping Robots

In parallel, the team has developed a one-legged humanoid robot that achieves repetitive jumps and 360-degree somersaults. These experiments were conducted without biological reference models, relying on a proprietary AI controller based on center-of-mass velocity and landing impact reduction.

Presentations and publications

The results were presented at the international conference “Humanoids 2025” and the “Conference on Robot Learning (CoRL)”, highlighting both hardware development and artificial intelligence achievements. Researchers JongHun Choe, Dongyun Kang and Gijeong Kim have been instrumental in these advances.

The project has been funded by the South Korean Ministry of Trade, Industry and Energy and the Korea Evaluation Institute of Technology (KEIT), cementing Korea as a leader in the development of autonomous humanoid robots for extreme environments.