Driving Like a Human: Imitation Learning for Path Planning using Convolutional Neural Networks

Human-like path planning is still a challenging task for automated vehicles. Imitation learning can teach these vehicles to learn planning from human demonstration. In this work, we propose to formulate the planning stage as a convolutional neural network (CNN). Thus, we can employ well established CNN techniques to learn planning from imitation. With the proposed method, we train a network for planning in complex traffic situations from both simulated and real world data. The resulting planning network exhibits human-like path generation.