Route Optimization in Service of a Search and Rescue Artificial Social Intelligence Agent

Search and Rescue (SAR) teams really rely on finding the best route to carry out their missions. In other words, it’s crucial for them to know the best ways to go in order to rescue people. This makes an Artificial Social Intelligence (ASI) tool, which knows the best routes available, very useful for a SAR team. This tool should know the environment really well and understand how the SAR team operates.

Imagine if this ASI tool could consider all this information while figuring out the best route, similar to how your GPS navigation app suggests the fastest route to your destination based on real-time traffic. However, until recently, it was tough for these ASI tools to do this in real-time because the calculations were too complex and took too much time.

But recent improvements in technology, specifically Graph Neural Networks, transformers, and attention models, could change that. These technologies could be used to quickly figure out nearly the best routes.

In this paper, we use these new tech tools as part of a decision-making framework to find routes for the teams participating in a search and rescue task organized by DARPA ASIST in the game Minecraft.

My Contribution

• > Developped a semantic graph environment for reinforcement learning and visulization

• > Immplemented 5 reinforcement learning and linear programming algorithms for route optimization.

• > Developed a sequential decision making framework leveraging Graph Transformer and the REINFORCE algorithm