This is an old revision of the document!


In this paper, we propose a new benchmark, called NoCrash, and perform a large scale analysis of end-toend behavioral cloning systems in complex driving conditions not studied in this context before. We use a high fidelity simulated environment based on the open source CARLA simulator [14] to enable reproducible large scale off-line training and on-line evaluation in over 80 hours of driving under several different conditions. We describe a strong Conditional Imitation Learning baseline, derived from [12], that significantly improves upon state-of-theart modular [26], affordance based [37], and reinforcement learning [27] approaches, both in terms of generalization performance in training environments and unseen ones.

Navigation