Dec-SGTS: Decentralized Sub-Goal Tree Search for Multi-Agent Coordination

• DOI: 10.1609/aaai.v35i13.17345
• Publication Date: 2022-09-08T19:49:53Z
• Authors: Minglong Li; Zhongxuan Cai; Wenjing Yang; Lixia Wu; Yinghui Xu; Ji Wang
• Keywords: Tree (set theory); Abstraction; Modularity

Multi-agent coordination tends to benefit from efficient communication, where cooperation often happens based on exchanging information about what the agents intend do, i.e. intention sharing. It becomes a key problem model by some proper abstraction. Currently, it is either too coarse such as final goals or fined primitive steps, which inefficient due lack of modularity and semantics. In this paper, we design novel multi-agent protocol subgoal intentions, defined probability distribution over feasible sequences. The intentions encode macro-action behaviors with so facilitate joint decision making at higher Built proposed protocol, present Dec-SGTS (Decentralized Sub-Goal Tree Search) solve decentralized online planning hierarchically efficiently. Each agent runs asynchronously iteratively performing three phases including local sub-goal tree search, update global We conduct experiments courier dispatching problem, results show that achieves much better reward while enjoying significant reduction time communication cost compared Dec-MCTS Monte Carlo Search).