Japanese researchers teach robots to swing like monkeys
Dual-arm robot masters monkey-bar brachiation with AI and fails gracefully
Researchers from the University of Tokyo’s Intelligent Systems and Informatics (IIS) lab have developed a life-sized dual-arm robot capable of brachiation—arm-based swinging locomotion similar to how gibbons and monkeys traverse trees. Their approach, called Waypoint-Guided Reinforcement Learning (WGRL), combines sparse waypoint guidance with reinforcement learning to generate non-linear whole-body motions. This method avoids the need for dense imitation data by specifying only key trajectory points, while a physics-informed reward structure (task success + energy efficiency) ensures stable and energy-conscious movement.
The team validated their system through both simulation and physical hardware experiments. In sim-to-sim tests with varied monkey-bar geometries, the robot maintained robust progression and recovered from failures like missed grips. Hardware trials confirmed these capabilities on real 3D-printed robot arms. The work, accepted to IROS 2026, provides a scalable blueprint for arm-based robotic locomotion and expands the operational workspace of humanoid robots in unstructured environments.
- WGRL uses sparse waypoints + RL to train robots without dense imitation data
- Robot achieves robust brachiation with forward motion, stability, and failure recovery
- Validated via sim-to-sim and hardware experiments on life-sized dual-arm robot
Why It Matters
Paves the way for robots to navigate complex, human-designed environments using purely arm-based locomotion—critical for search, repair, and exploration tasks.