Adaptive and robust deep reinforcement learning for drone systems

About the Project

The increasing complexity of multi-agent systems in robotics has amplified the need for intelligent, cooperative decision-making among agents. Deep reinforcement learning is a promising approach for training agents to learn optimal strategies in cooperative, competitive, or mixed environments. However, it is often inherently challenging due to issues such as non-stationarity, partial observability, and credit assignment, making the development of effective learning algorithms a critical area of research. Existing approaches have shown progress in addressing these issues and enhancing the agents' ability to interpret limited perceptions of their environment. Nevertheless, these complexities can significantly hinder performance, particularly in environments characterised by dynamic interactions and uncertainties.

This project will focus on leveraging multi-agent deep reinforcement learning techniques to address these complex challenges in the context of drone systems. As such systems become increasingly integral to applications ranging from environmental monitoring to disaster response, the ability to effectively coordinate multiple drone agents in real-time is critical. The research will explore novel strategies to enhance the robustness and adaptability of drone operations, ensuring that agents can operate smoothly and efficiently in diverse conditions, contributing significantly to the advancement of autonomous drone systems.

The project will be conducted on simulated environments, but may incorporate aspects of sim-to-real transfer, bridging the gap between simulation and real-world applications.

Funding Notes

there is no funding for this project