Deer in The Headlights: Short Term Planning via Reinforcement Learning Algorithms for Autonomous Vehicles

Reinforcement learning has seen a resurgence recently in tandem with the resurgence of deep neural networks. Recent work done by DeepMind and others has shown us the enormous potential of reinforcement learning when applied to certain problem domains. Of these domains, one of them is of particular contemporary relevance; Autonomous Vehicles. Several multinational corporations are pouring enormous resources into developing cars which can autonomously drive themselves. Autonomous Vehicles are expected to bring about the largest revolution in the automobile industry, since the invention of the internal combustion engine. One of the key components of autonomous vehicles is reinforcement learning. In this paper, we explored the performance of a Reinforcement Learning agent, known as Dueling DQN, to a simulated driving environment, presenting a short term planner to be used as a lane following and obstacle avoidance system as part of a larger, integrated system.