Adaptive Task Token Framework for Multi-Agent Logistics Operations

doi:10.1109/lra.2026.3732893

Adaptive Task Token Framework for Multi-Agent Logistics Operations

2026-09-10

0

PRE

AI

DOI：10.1109/lra.2026.3732893

Original

Original request for help

Save

Abstract

En 中文

Multi-Agent Pickup and Delivery (MAPD) extends Multi-Agent Path Finding to long-horizon settings where pickupand- delivery tasks arrive online and agents must be assigned and routed continuously, as in automated warehouses. Existing MAPD methods leave three gaps that limit real-world deployment. First, token-based methods such as Token Passing (TP) and TP with Task Swaps (TPTS) assign tasks sequentially, yielding greedy, globally suboptimal allocations and leaving idle agents parked where they obstruct active agents. Second, these methods assume disruption-free execution and provide no recovery from the delays and deadlocks common on real robots; k-TP adds delay tolerance but only conservatively and at high computational cost. Third, centralized optimal methods such as CENTRAL coordinate well but do not scale to realtime operation. To close these gaps, we propose the Adaptive Task Token Framework (ATTF), which (i) replaces sequential assignment with a global, proximity-based module that pairs all idle agents with all available tasks simultaneously by travel cost, and (ii) introduces a unified delay-and-deadlock-resilient mechanism that truncates and replans only affected paths and redirects idle agents to non-task endpoints to relieve congestion. Across diverse warehouse layouts, ATTF matches or improves on TP and TPTS in makespan and service time, and outperforms k-TP on all three metrics under delays - reducing service time by up to 49% and runtime by $\sim$82% - while sustaining sub-second per-timestep runtime for up to 500 agents, a scale at which k-TP fails to produce solutions.

Keywords:

Timing

Delays

Warehousing

Algorithms

Runtime

Planning

Conferences

Real-time systems

Path planning

Radio access networks

Adaptive Task Token Framework for Multi-Agent Logistics Operations

Abstract

Journal

Organization

Cited Papers