Together We Rise: Optimizing Real-Time Multi-Robot Task Allocation using Coordinated Heterogeneous Plays

TL;DR

The paper tackles real-time multi-robot task allocation (MRTA) in dynamic warehouses, where tasks arrive with start and end locations and must be assigned to robots under time and safety constraints. It introduces MRTAgent, a self-play inspired bi-level reinforcement learning framework with a Planner for task selection and an Executor for robot allocation, augmented by a LQR-based Navigator with artificial potential field collision avoidance. Key contributions include integrating task selection, robot assignment, and physics-aware navigation within a single end-to-end framework, handling robot dynamics and state-of-charge (SOC) constraints, and validating robustness to distributional shifts. The approach demonstrates superior performance over baselines on synthetic datasets, with strong implications for improving real-time efficiency in online order fulfillment and scalable operation across varying fleet sizes and task volumes.

Abstract

Efficient task allocation among multiple robots is crucial for optimizing productivity in modern warehouses, particularly in response to the increasing demands of online order fulfillment. This paper addresses the real-time multi-robot task allocation (MRTA) problem in dynamic warehouse environments, where tasks emerge with specified start and end locations. The objective is to minimize both the total travel distance of robots and delays in task completion, while also considering practical constraints such as battery management and collision avoidance. We introduce MRTAgent, a dual-agent Reinforcement Learning (RL) framework inspired by self-play, designed to optimize task assignments and robot selection to ensure timely task execution. For safe navigation, a modified linear quadratic controller (LQR) approach is employed. To the best of our knowledge, MRTAgent is the first framework to address all critical aspects of practical MRTA problems while supporting continuous robot movements.

Figures (3)

• Figure 1: Self-play inspired bi-level reinforcement learning agent for assigning robots to tasks within a dynamic environment
• Figure 2: Self-Play Motivated Bi-level RL Agent Architecture with trainable weights denoted by $\{W_k, W_k^{\prime}\}$
• Figure 3: Learning curves of the MRTAgent