Learning Decentralized Multi-Biped Control for Payload Transport
Published in 8th Conference on Robot Learning (CoRL) 2024
Recommended citation: B. Pandit, A. Gupta, M. Gadde, A. Johnson, A. Shrestha, H. Duan, J. Dao and A. Fern, ”Learning Decentralized Multi-Biped Control for Payload Transport,” Proceedings of The 8th Conference on Robot Learning, Munich, Germany, 2024
The paper introduces a decentralized control system for multi-biped robot carriers designed to transport payloads over rough terrain, where legs are more effective than wheels. The approach uses multiple bipedal robots rigidly attached to a carrier, with a controller trained via reinforcement learning in simulation, and then transferred to real-world applications. The system is scalable, adapting to varying numbers and configurations of bipedal robots without the need for retraining. The effectiveness of the controller is demonstrated through both simulated transport scenarios and real-world experiments using Cassie robots.
Our main contributions are:
- Decentralized control system for multi-biped robot carriers, allowing payload transport over rough terrain.
- Reinforcement learning-based controller that adapts to different configurations of bipedal robots without retraining.
- Scalable system capable of handling varying numbers of bipedal robots in a flexible and effective manner.
- Real-world validation through experiments with two and three Cassie robots, marking the first scalable multi-biped transport system.
