.png){kind=icon}
Return
Collaborative-Online-Learning-enabled distributionally robust motion control for multi-robot systems
.png){kind=icon}
PRE
AI
DOI：10.1016/j.automatica.2026.113102.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper proposes a novel COllaborative-Online-Learning (COOL)-enabled motion control framework for multi-robot systems to avoid collisions with randomly moving obstacles whose motion distributions are partially observable via decentralized data streams. To address data acquisition challenges caused by occlusion, a COOL scheme based on the Dirichlet process mixture model is developed to efficiently extract distribution information through the exchange of selected learning structures among robots. Leveraging the learned local-moment information, a data-driven ambiguity set for obstacle motion is constructed and propagated over the entire prediction horizon using current obstacle positions. To balance control performance and computational efficiency, a compression scheme with safety guarantees is introduced to adaptively regulate the complexity of the ambiguity set. Probabilistic collision-free trajectories are then generated via distributionally robust optimization, where obstacle avoidance constraints are equivalently reformulated by tractable semi-definite programs. Finally, probabilistic collision avoidance and long-term tracking performance guarantees are established, and numerical simulations validate the superiority of the proposed framework. (c) 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
Keywords:
Collaborative online learning
Distributionally robust motion control
Separation hyperplane
Multi-robot system
Journal
IF:
5.9
Papers:
1.2W
Citations:
5.2W
Organization
Cited Papers
No cited papers available