Event
Zhejiang University proposes GuardPIBT, enhancing PIBT planning with residual re-ranking
1 reports · 1 sources
A team at Zhejiang University proposed GuardPIBT for large-scale three-dimensional multi-agent path finding (MAPF): the neural network only performs residual re-ranking proposals on PIBT's native candidate actions, while conflict checking, priority inheritance, and backtracking are still handled by the PIBT core, and counterfactual grouped gating is used to filter harmful interventions; in simulation, it completed conflict-free three-dimensional planning for 1 million agents. All results are simulations of MAPF on discrete graphs and have not yet been validated outside a simulation environment; in small-scale scenarios, the neural network introduces additional inference overhead, making it slower than the pure search-based LaCAM, and asynchronous cached inference also introduces latency.
Report timeline
- Zhejiang University's GuardPIBT enhances PIBT with residual reranking, completes conflict-free 3D planning for 1 million agents in simulation
A Zhejiang University team proposed GuardPIBT, which, while retaining the PIBT execution kernel, uses a neural network to make residual reranking proposals over PIBT's native candidate actions; conflict checking, priority inheritance, and backtracking are still handled by the PIBT kernel, and counterfactual grouped gating is introduced to filter harmful interventions.
深蓝AI · 微信公众号 ·