Natural disasters and man-made emergencies often pose significant challenges to search and rescue (SAR) operations, leading to delays in locating survivors and increasing casualties. Unmanned aerial vehicles (UAVs), also known as drones, offer promising potential for improving SAR efficiency due to their maneuverability, aerial perspective, and ability to operate in hazardous environments. However, effectively coordinating multiple UAVs in complex, dynamic SAR situations remains a significant challenge.
This project proposes the development and evaluation of a multi-agent reinforcement learning (MARL) framework for optimizing UAV search and rescue operations. By enabling autonomous decision-making and collaboration among UAVs, this approach aims to significantly enhance search coverage, reduce search time, and ultimately save more lives.
Develop a MARL framework for multi-UAV collaboration in SAR scenarios, including:
- Environment representation: Design a realistic simulation environment that captures key SAR challenges, such as dynamic obstacles, uneven terrain, and limited communication channels.
- Agent design: Implement individual UAV agents with sensors and actuators, capable of learning and adapting their actions based on observations and rewards.
- Communication protocols: Design efficient communication mechanisms for information sharing and coordination among agents, considering bandwidth limitations and potential disruptions.
- Reward function: Define a reward function that incentivizes efficient search coverage, timely survivor detection, and safe operation within the environment.
- Train the MARL system using reinforcement learning algorithms, such as Deep Q-Network (DQN) or Multi-Agent Proximal Policy Optimization (MAPPO).
- Evaluate the performance of the trained MARL system in simulated SAR scenarios, comparing it to baseline approaches like pre-programmed search patterns or simple centralized control.
- Analyze the learned behavior of the agents to understand how they collaborate and make decisions in different situations.
- A robust and scalable MARL framework for multi-UAV SAR operations.
- Improved search efficiency and reduced search time compared to traditional approaches.
- Increased probability of survivor detection and reduced risk of harm to UAVs.
- Insights into the decision-making process of collaborative agents through analysis of their learned behavior.