IndoorExplorers: an OpenAI Gym environment for Multi-UAV Exploration Algorithms

11/27/2023

The goal of this work was to create an OpenAI Gym environment to simulate indoor exploration scenarios by a swarm of autonomous Unmanned Aerial Vehicles (UAVs), each equipped with Light Detection And Ranging (LiDAR) sensors and with safe flying capabilities, including the detection and avoidance of any objects, across the space in question. The exploration tasks consists in determining the optimal path that gathers as much information about the space as possible, in this case to create a map of the space. Using a swarm of UAVs, it is possible to achieve these tasks faster, with fewer costs and safely for humans. The developed OpenAI Gym-based environment was then used to test a Reinforcement Learning (RL) algorithm for path planning, specifically Dueling Double Deep Q-Learning (DDDQN). The developed environment currently allows tests in 2D maps with up to four UAVs equipped with a simplified simulated LiDAR sensor, with or without communications. The results obtained compare two approaches to accelerate the training of the DDDQN. Furthermore, an analysis of the impact of more than one agent and whether communications affect the performance was done.

Authors in the community:

• 

  A

  Alexandra Isabel Fernandes

  ist190001

Supervisors of this institution:

• 

  A

  António Manuel Raminhos Cordeiro Grilo

  ist14017

• 

  J

  João Paulo Baptista de Carvalho

  ist14039

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico