WARL: Wrench-Augmented Reinforcement Learning for Task-Agnostic Learning in Legged Robots
Proceedings of the 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026
Master's student
Keita Yoneda
JSK Robotics Laboratory · The University of Tokyo
From mechanism to learning — building quadruped robots that keep getting better out in the real world.
About
I work across the whole stack of a legged robot — designing the mechanism, writing the software that drives it, and integrating the two into a machine that leaves the bench — with machine learning as my specialty. My recent work centres on KLEIYN, a quadruped with an active waist joint that walks, climbs chimneys and stands up on two legs. I am currently most interested in giving robots new sensing modalities, in continual learning that lets them keep growing after deployment, and in putting quadrupeds to work in the real world.
- Quadruped Robots
- Reinforcement Learning
- Continual Learning
- Multi-modal Sensing
- Bioinspired Mechanism Design
- Real-World Deployment
Selected Publications
All publications
EFGCL: Learning Dynamic Motion Through Spotting-Inspired External Force Guided Curriculum Learning
IEEE Robotics and Automation Letters (RA-L), 2026
KLEIYN: A Quadruped Robot with an Active Waist for Both Locomotion and Wall Climbing
Proceedings of the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025
Awards
-
2025
IEEE Robotics and Automation Society Japan Joint Chapter Young Award
for “KLEIYN: A Quadruped Robot with an Active Waist for Both Locomotion and Wall Climbing”
-
2025
Workshop on Climbing Robotics Best Early-Career Talk Award
for “Chimney Climbing of Quadruped Robot KLEIYN With Active Waist Joint Based on Contact-Guided Curriculum Learning”