# UAV Pathfinding in Dynamic Obstacle Avoidance with Multi-agent Reinforcement Learning

**Authors:** Qizhen Wu, Lei Chen, Kexin Liu, Jinhu Lv  
**arXiv:** [2310.16659](https://arxiv.org/abs/2310.16659)  
**Date:** 2023-10-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2310.16659v1/

## Key points

- Paper proposes a novel approach for online planning of feasible and safe paths for agents in dynamic and uncertain scenarios
- Introduces centralized training with decentralized execution method based on multi-agent reinforcement learning
- Agents communicate with central planner or neighbors to plan paths online
- Incorporates model predictive control to improve training efficiency and sample utilization
- Validated through experiments in simulation, indoor, and outdoor environments
- Results demonstrate successful dynamic obstacle avoidance and provision of feasible and safe paths for UAVs
- Video demonstration enhances understanding of the proposed method
- Promising solution for pathfinding in dynamic obstacle avoidance using multi-agent reinforcement learning

## Summary

The paper titled "UAV Pathfinding in Dynamic Obstacle Avoidance with Multi-agent Reinforcement Learning" proposes a novel approach to address the problem of online planning of feasible and safe paths for agents in dynamic and uncertain scenarios. To overcome the limitations of existing methods such as fully centralized and fully decentralized approaches, which face challenges such as dimension explosion and poor convergence, the authors introduce a centralized training with decentralized execution method based on multi-agent reinforcement learning. In this approach, each agent communicates either with the central planner or its neighbors to plan feasible and safe paths online. Moreover, the concept of model predictive control is incorporated to improve training efficiency and sample utilization of agents. The effectiveness of the proposed method is validated through experiments conducted in simulation, indoor, and outdoor environments. The results demonstrate that the approach successfully addresses the dynamic obstacle avoidance problem and provides feasible and safe paths for UAVs. Furthermore, a video demonstration further enhances understanding of the proposed method. Overall, this paper presents a promising solution for pathfinding in dynamic obstacle avoidance using multi-agent reinforcement learning.