# Multi-vehicle Flocking Control with Deep Deterministic Policy Gradient Method

arXiv: https://arxiv.org/abs/1806.00196 · HTML: https://mlsanity.com/paper/1806.00196/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Jun 2018.

## Built on by (most cited first, 7 of 7)

- [Context-Aware Sparse Deep Coordination Graphs](https://mlsanity.com/paper/2106.02886/) (2021)
- [Hokoff: Real Game Dataset from Honor of Kings and its Offline Reinforcement Learning Benchmarks](https://mlsanity.com/paper/2408.10556/) (2024)
- [Non-Linear Coordination Graphs](https://mlsanity.com/paper/2211.08404/) (2022)
- [Autonomous Unmanned Aerial Vehicle Navigation using Reinforcement Learning: A Systematic Review](https://mlsanity.com/paper/2208.12328/) (2022)
- [QTypeMix: Enhancing Multi-Agent Cooperative Strategies through Heterogeneous and Homogeneous Value Decomposition](https://mlsanity.com/paper/2408.07098/) (2024)
- [Sample-Efficient Multi-Agent Reinforcement Learning with Demonstrations for Flocking Control](https://mlsanity.com/paper/2209.08351/) (2022)
- [AIR: Unifying Individual and Collective Exploration in Cooperative Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/2412.15700/) (2024)

## Abstract (arXiv)

Flocking control has been studied extensively along with the wide application of multi-vehicle systems. In this paper the Multi-vehicles System (MVS) flocking control with collision avoidance and communication preserving is considered based on the deep reinforcement learning framework. Specifically the deep deterministic policy gradient (DDPG) with centralized training and distributed execution process is implemented to obtain the flocking control policy. First, to avoid the dynamically changed observation of state, a three layers tensor based representation of the observation is used so that the state remains constant although the observation dimension is changing. A reward function is designed to guide the way-points tracking, collision avoidance and communication preserving. The reward function is augmented by introducing the local reward function of neighbors. Finally, a centralized training process which trains the shared policy based on common training set among all agents. The proposed method is tested under simulated scenarios with different setup.