Multi-agent Route Planning using Deep Reinforcement Learning Techniques and Transformer Networks for Graph Analysis

Today, November 4th, Daniel Fuertes has defended his Ph.D thesis titled: “Multi-agent Route Planning using Deep Reinforcement Learning Techniques and Transformer Networks for Graph Analysis.”

In his work, Daniel addresses one of today’s most significant challenges: autonomous navigation in complex environments, with applications ranging from package delivery to rescue operations. Traditionally, these systems have relied on human operators, especially in multi-agent scenarios where coordination and collaboration are essential.

To tackle these challenges, his research introduces three innovative models based on Transformer neural networks and deep reinforcement learning:

1. 1. 1. FCM-Transformer:
         • A two-phase approach that clusters regions and assigns routes to each agent.
         • Introduces a region-sharing strategy, promoting cooperation and overcoming limitations of traditional clustering methods.
      2. NaviFormer:
         • Evolves the FCM-Transformer by integrating route and trajectory planning, solving waypoint sequencing and path planning as a single problem.
         • Enables the prediction of precise, collision-free trajectories, eliminating the need for separate solutions and increasing efficiency.
      3. TOP-Former:
         • Directly addresses the Team Orienteering Problem for multiple agents.
         • Considers the global state of all agents simultaneously, ensuring robust coordination and high-quality routes in complex scenarios.

Experimental results show that these models achieve an exceptional balance between solution quality and computational efficiency, remaining highly robust under various conditions. However, challenges such as scalability in large-scale problems remain open, highlighting areas for future research.

With this Ph.D thesis, Daniel Fuertes makes a significant contribution to the field of artificial intelligence applied to multi-agent autonomous navigation, providing promising tools for the efficient coordination of autonomous vehicles in complex scenarios.