Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DRAL: Deep Reinforcement Adaptive Learning for Multi-UAVs Navigation in Unknown Indoor Environment

Kangtong Mo, Linyue Chu, Xingyu Zhang, Xiran Su, Yang Qian, Yining Ou, Wian Pretorius

TL;DR

This work tackles indoor UAV navigation under GPS-denied conditions and unknown payloads by introducing DRAL, a deep reinforcement adaptive learning framework that combines a physics-based multi-UAV payload model with a learning-based controller trained in simulation. The method uses PPO with an Asymmetric Actor-Critic to exploit privileged training information and achieve zero-shot transfer to real environments via domain randomization, enabling coordinated payload transport with multiple drones. Key contributions include a detailed dual-UAV–payload dynamics model, a continuous-action DRAL design, and extensive simulation benchmarking showing improved success rates and faster, smoother trajectories over traditional RL baselines. The findings demonstrate robust, adaptive multi-drone capabilities for confined indoor operations, with clear potential for real-world deployment in tasks like unknown-payload retrieval and collaborative lifting.

Abstract

Autonomous indoor navigation of UAVs presents numerous challenges, primarily due to the limited precision of GPS in enclosed environments. Additionally, UAVs' limited capacity to carry heavy or power-intensive sensors, such as overheight packages, exacerbates the difficulty of achieving autonomous navigation indoors. This paper introduces an advanced system in which a drone autonomously navigates indoor spaces to locate a specific target, such as an unknown Amazon package, using only a single camera. Employing a deep learning approach, a deep reinforcement adaptive learning algorithm is trained to develop a control strategy that emulates the decision-making process of an expert pilot. We demonstrate the efficacy of our system through real-time simulations conducted in various indoor settings. We apply multiple visualization techniques to gain deeper insights into our trained network. Furthermore, we extend our approach to include an adaptive control algorithm for coordinating multiple drones to lift an object in an indoor environment collaboratively. Integrating our DRAL algorithm enables multiple UAVs to learn optimal control strategies that adapt to dynamic conditions and uncertainties. This innovation enhances the robustness and flexibility of indoor navigation and opens new possibilities for complex multi-drone operations in confined spaces. The proposed framework highlights significant advancements in adaptive control and deep reinforcement learning, offering robust solutions for complex multi-agent systems in real-world applications.

DRAL: Deep Reinforcement Adaptive Learning for Multi-UAVs Navigation in Unknown Indoor Environment

TL;DR

This work tackles indoor UAV navigation under GPS-denied conditions and unknown payloads by introducing DRAL, a deep reinforcement adaptive learning framework that combines a physics-based multi-UAV payload model with a learning-based controller trained in simulation. The method uses PPO with an Asymmetric Actor-Critic to exploit privileged training information and achieve zero-shot transfer to real environments via domain randomization, enabling coordinated payload transport with multiple drones. Key contributions include a detailed dual-UAV–payload dynamics model, a continuous-action DRAL design, and extensive simulation benchmarking showing improved success rates and faster, smoother trajectories over traditional RL baselines. The findings demonstrate robust, adaptive multi-drone capabilities for confined indoor operations, with clear potential for real-world deployment in tasks like unknown-payload retrieval and collaborative lifting.

Abstract

Autonomous indoor navigation of UAVs presents numerous challenges, primarily due to the limited precision of GPS in enclosed environments. Additionally, UAVs' limited capacity to carry heavy or power-intensive sensors, such as overheight packages, exacerbates the difficulty of achieving autonomous navigation indoors. This paper introduces an advanced system in which a drone autonomously navigates indoor spaces to locate a specific target, such as an unknown Amazon package, using only a single camera. Employing a deep learning approach, a deep reinforcement adaptive learning algorithm is trained to develop a control strategy that emulates the decision-making process of an expert pilot. We demonstrate the efficacy of our system through real-time simulations conducted in various indoor settings. We apply multiple visualization techniques to gain deeper insights into our trained network. Furthermore, we extend our approach to include an adaptive control algorithm for coordinating multiple drones to lift an object in an indoor environment collaboratively. Integrating our DRAL algorithm enables multiple UAVs to learn optimal control strategies that adapt to dynamic conditions and uncertainties. This innovation enhances the robustness and flexibility of indoor navigation and opens new possibilities for complex multi-drone operations in confined spaces. The proposed framework highlights significant advancements in adaptive control and deep reinforcement learning, offering robust solutions for complex multi-agent systems in real-world applications.
Paper Structure (7 sections, 4 equations, 4 figures, 1 table)

This paper contains 7 sections, 4 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related work
  3. Methodology
  4. Multi-drone Dynamics
  5. Deep reinforcement adaptive learning Design
  6. Experiment Results
  7. Conclusion

Figures (4)

  • Figure 1: A multi-drone system using cables to lift an unknown payload to transport it during the indoor navigation.
  • Figure 2: Simulation environment setup in Gazebo.
  • Figure 3: Three quadrotors elevate an unidentified payload during an autonomous exploration assessment executed within a multifaceted indoor environment.
  • Figure 4: The investigative advancement of four methodologies. The results show that our method is better than the other three popular deep learning methods.