A Dynamic Risk-aware Routing Recommendation Using Deep Reinforcement Learning

Crime risks in urban environments vary over time and space, making it challenging to find a safe and efficient route. Previous solutions to this problem were too simplistic and static, relying on exhaustive searches that were not practical. In this paper, we present a new approach using deep reinforcement learning for dynamic risk-aware path planning. Our method evaluates crime risks and distances to suggest optimal routes that adapt to changing conditions. We have designed a reward system that considers road length, risk levels, and the decay of risks over time. This innovative approach allows our model to make decisions that balance distance and crime risk. Through extensive experiments, we demonstrate the effectiveness of our model compared to traditional path-planning methods, especially as the road network grows in size.