Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MARLander: A Local Path Planning for Drone Swarms using Multiagent Deep Reinforcement Learning

Demetros Aschu, Robinroy Peter, Sausar Karaf, Aleksey Fedoseev, Dzmitry Tsetserukou

TL;DR

MARLander tackles the challenge of safe, precise swarm landings by learning decentralized control policies via multi-agent deep reinforcement learning. The method uses a PPO-based policy that relies on local observations from a two-drone system, trained in a PyBullet-based simulation with random 4×4×4 m setups and deployed on Crazyflie drones with Vicon indoor localization. Results show cm-scale landing accuracy on stationary targets (≈2.26–2.97 cm) and ≈3.93 cm on moving platforms, with high success rates (≈91.67% static, ≈75% moving), outperforming PID+APF baselines and single-agent RL. The work demonstrates scalable, decentralized coordination for drone swarms with potential impact on logistics, safety, and rescue missions.

Abstract

Achieving safe and precise landings for a swarm of drones poses a significant challenge, primarily attributed to conventional control and planning methods. This paper presents the implementation of multi-agent deep reinforcement learning (MADRL) techniques for the precise landing of a drone swarm at relocated target locations. The system is trained in a realistic simulated environment with a maximum velocity of 3 m/s in training spaces of 4 x 4 x 4 m and deployed utilizing Crazyflie drones with a Vicon indoor localization system. The experimental results revealed that the proposed approach achieved a landing accuracy of 2.26 cm on stationary and 3.93 cm on moving platforms surpassing a baseline method used with a Proportional-integral-derivative (PID) controller with an Artificial Potential Field (APF). This research highlights drone landing technologies that eliminate the need for analytical centralized systems, potentially offering scalability and revolutionizing applications in logistics, safety, and rescue missions.

MARLander: A Local Path Planning for Drone Swarms using Multiagent Deep Reinforcement Learning

TL;DR

MARLander tackles the challenge of safe, precise swarm landings by learning decentralized control policies via multi-agent deep reinforcement learning. The method uses a PPO-based policy that relies on local observations from a two-drone system, trained in a PyBullet-based simulation with random 4×4×4 m setups and deployed on Crazyflie drones with Vicon indoor localization. Results show cm-scale landing accuracy on stationary targets (≈2.26–2.97 cm) and ≈3.93 cm on moving platforms, with high success rates (≈91.67% static, ≈75% moving), outperforming PID+APF baselines and single-agent RL. The work demonstrates scalable, decentralized coordination for drone swarms with potential impact on logistics, safety, and rescue missions.

Abstract

Achieving safe and precise landings for a swarm of drones poses a significant challenge, primarily attributed to conventional control and planning methods. This paper presents the implementation of multi-agent deep reinforcement learning (MADRL) techniques for the precise landing of a drone swarm at relocated target locations. The system is trained in a realistic simulated environment with a maximum velocity of 3 m/s in training spaces of 4 x 4 x 4 m and deployed utilizing Crazyflie drones with a Vicon indoor localization system. The experimental results revealed that the proposed approach achieved a landing accuracy of 2.26 cm on stationary and 3.93 cm on moving platforms surpassing a baseline method used with a Proportional-integral-derivative (PID) controller with an Artificial Potential Field (APF). This research highlights drone landing technologies that eliminate the need for analytical centralized systems, potentially offering scalability and revolutionizing applications in logistics, safety, and rescue missions.
Paper Structure (26 sections, 6 equations, 8 figures, 2 tables)

This paper contains 26 sections, 6 equations, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. System Overview
  5. Problem Formulation and Preliminaries
  6. Environment Setup
  7. Observation and Actions
  8. Reward Function
  9. Model Architecture
  10. Simulation Setup
  11. Training Configuration
  12. Experiments
  13. Experimental Setup
  14. MARLander experiment with stationary platform
  15. Description
  16. ...and 11 more sections

Figures (8)

  • Figure 1: MARlander two drones landing on the target platform placed on the robot manipulator
  • Figure 2: General system overview of MARLander.
  • Figure 3: A neural network architecture for PPO algorithm
  • Figure 4: Gym PyBullet environment for simulating a MADRL-driven swarm of drones
  • Figure 5: Episode reward mean during training
  • ...and 3 more figures