Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Combined Problem of Online Task Assignment and Lifelong Path Finding in Logistics Warehouses: Rule-Based Systems Matter

Fengming Zhu, Weijia Xu, Yifei Guo, Fangzhen Lin

TL;DR

The paper tackles online Task Assignment and Lifelong Path Finding (TAPF) in automated warehouses, introducing a formal framework for joint decision-making under the Type$\odot$ robot model. It contributes a rule-based lifelong path planner, Touring With Early Exit, and an online assignment approach that includes stateless, adaptive, and predictive (MDP/RL) methods, with PPO used to learn a delivery policy. Empirical results on Meituan-scale warehouse simulations show substantial improvements: execution time drops to $83.77\%$ of the current system while maintaining throughput with nontrivial reductions in required agents (down to $60\%$). The work highlights the importance of layout-aware planning and provides a practical, scalable approach with open avenues for integrating routing and assignment more tightly in real-world deployments.

Abstract

We study the combined problem of online task assignment and lifelong path finding, which is crucial for the logistics industries. However, most literature either (1) focuses on lifelong path finding assuming a given task assigner, or (2) studies the offline version of this problem where tasks are known in advance. We argue that, to maximize the system throughput, the online version that integrates these two components should be tackled directly. To this end, we introduce a formal framework of the combined problem and its solution concept. Then, we design a rule-based lifelong planner under a practical robot model that works well even in environments with severe local congestion. Upon that, we automate the search for the task assigner with respect to the underlying path planner. Simulation experiments conducted in warehouse scenarios at Meituan, one of the largest shopping platforms in China, demonstrate that (a)in terms of time efficiency, our system requires only 83.77% of the execution time needed for the currently deployed system at Meituan, outperforming other SOTA algorithms by 8.09%; (b)in terms of economic efficiency, ours can achieve the same throughput with only 60% of the agents currently in use. The code and demos are available at https://github.com/Fernadoo/Online-TAPF.

The Combined Problem of Online Task Assignment and Lifelong Path Finding in Logistics Warehouses: Rule-Based Systems Matter

TL;DR

The paper tackles online Task Assignment and Lifelong Path Finding (TAPF) in automated warehouses, introducing a formal framework for joint decision-making under the Type robot model. It contributes a rule-based lifelong path planner, Touring With Early Exit, and an online assignment approach that includes stateless, adaptive, and predictive (MDP/RL) methods, with PPO used to learn a delivery policy. Empirical results on Meituan-scale warehouse simulations show substantial improvements: execution time drops to of the current system while maintaining throughput with nontrivial reductions in required agents (down to ). The work highlights the importance of layout-aware planning and provides a practical, scalable approach with open avenues for integrating routing and assignment more tightly in real-world deployments.

Abstract

We study the combined problem of online task assignment and lifelong path finding, which is crucial for the logistics industries. However, most literature either (1) focuses on lifelong path finding assuming a given task assigner, or (2) studies the offline version of this problem where tasks are known in advance. We argue that, to maximize the system throughput, the online version that integrates these two components should be tackled directly. To this end, we introduce a formal framework of the combined problem and its solution concept. Then, we design a rule-based lifelong planner under a practical robot model that works well even in environments with severe local congestion. Upon that, we automate the search for the task assigner with respect to the underlying path planner. Simulation experiments conducted in warehouse scenarios at Meituan, one of the largest shopping platforms in China, demonstrate that (a)in terms of time efficiency, our system requires only 83.77% of the execution time needed for the currently deployed system at Meituan, outperforming other SOTA algorithms by 8.09%; (b)in terms of economic efficiency, ours can achieve the same throughput with only 60% of the agents currently in use. The code and demos are available at https://github.com/Fernadoo/Online-TAPF.

Paper Structure

This paper contains 20 sections, 1 equation, 8 figures, 2 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Definition
  4. Path Finding
  5. Existing Lifelong Path Finding Algorithms
  6. Our Rule-Based Solution: Touring With Early Exit
  7. Comparison for Path Finding Algorithms
  8. Task Assignment
  9. Stateless Assignment.
  10. Adaptive Assignment.
  11. Predictive Assignment.
  12. Experimental Results
  13. Conclusion and Future Work
  14. Relation to MAPD
  15. More Discussion on Related Work
  16. ...and 5 more sections

Figures (8)

  • Figure 1: A non-well-formed instance currently deployed in Meituan warehouses. The white cells near Green dots are delivery ports, while the ones near Red dots are pickup ports. Colored circles heading to different directions with numbers are agents. The colored box (blue) is a pickup port currently assigned to the agent in the same color (ID 45 in the lower right area). Congestion happens a lot near the pickup ports.
  • Figure 2: Rule1-Touring (a) and two prioritized cases (b, c). Colored boxes are the goals.
  • Figure 3: The tested makespans of lifelong path finding algorithms in 50-agent Meituan warehouse scenarios. Dotted lines represent the 25-/75-quantiles, and white dots are the means. The means correspond to the leftmost column of the 50-agent scenario in Table \ref{['tab:eval_full']}. 416.09 is the reference makespan under Meituan's current system.
  • Figure 4: The dynamics of the occupation ratios for different stateless and adaptive assignment strategies in 50-agent cases. Dashed lines represent the means.
  • Figure 5: Examples showing the added difficulty of resolving collisions with the Type$\odot$ robot model.
  • ...and 3 more figures

Theorems & Definitions (3)

  • Definition 1: Feasibility
  • Definition 2: Makespan
  • Example 1: System pipeline