A ’Cognitive Driving Framework’ for Collision Avoidance in Autonomous Vehicles

Abstract

The Cognitive Driving Framework is a novel method for forecasting the future states of a multi-agent system that takes into consideration both the intentions of the agents as well as their beliefs about the environment. This is particularly useful for autonomous vehicles operating in an urban environment. The algorithm maintains a posterior probability distribution over agent intents and beliefs in order to more accurately forecast their future behavior. This allows an agent navigating the environment to recognize dangerous situations earlier and more accurately than competing algorithms, therefore allowing the agent take actions in order to prevent collisions. This paper presents the Cognitive Driving Framework in detail and describes its application to intersection navigation for autonomous vehicles. The effects of different parameter choices on the performance of the algorithm are analyzed and experiments are conducted demonstrating the ability of the algorithm to predict and prevent automobile collisions caused by human error in multiple intersection navigation scenarios. The results are compared to the performance of prevailing methods; namely reactionary planning and constant velocity forecasting.