EFFICIENT ROUTING WITH INVERSE REINFORCEMENT LEARNING

Main Article Content

Srikrishnan Subramanian
Adithya Raam Sankar

Abstract

Transportation has undergone a lot of evolution since the invention of the wheel. With more and more sophisti-cated manufacturing methods being implemented, the production time of new vehicles has reduced drastically. This has led to a substantial increase in vehicular traffic in the last 3 decades. The beginning of personalized transportation has ushered in a new dimension in the understanding of traffic. The initial approach in managing spatial areas was to merge the roads and visualize it as a graphed network, where every stretch of road is visualized as an edge and the digressions/splitting of traffic being nodes leading to other edges. Increased vehicular traffic with almost constant mapped spatial area causes an unstable equilibrium, leading to congestion and gridlocks. This equilibrium requires an effective balancing/routing strategy to maintain stability among the network. The correlation between road networks and com-puter networks has been exploited to solve this problem, expecting minimal deviation from ideal behavior. Networking protocols are unable to handle deviations that occur due to natural human behavior. Machine Learning Techniques can be implemented to understand these deviations and obtain patterns in real time. The proposed system approaches the routing problem with the aim of learning optimal reward functions by observing regular human behavior for a set of actions. These functions are pivotal in maximizing utility for every agent involved in the procedure by adopting a cooperative and interactive approach.

Downloads

Download data is not yet available.

Article Details

Section
Articles