Dynamic Conditional Imitation Learning for Autonomous Driving

Conditional imitation learning (CIL) trains deep neural networks, in an end-to-end manner, to mimic human driving. This approach has demonstrated suitable vehicle control when following roads, avoiding obstacles, or taking specific turns at intersections reach a destination. Unfortunately, performance dramatically decreases deployed unseen environments and is inconsistent against varying weather conditions. Most importantly, the current CIL fails avoid static road blockages. In this work, we propose solution those deficiencies. First, fuse laser scanner with regular camera streams, features level, overcome generalization consistency challenges. Second, introduce new efficient Occupancy Grid Mapping (OGM) method along algorithms for blockages avoidance global route planning. Consequently, our proposed dynamically detects partial full blockages, guides controlled another Following original effectiveness of proposal on CARLA simulator urban driving benchmark. Our experiments showed that model improved conditions by four times autonomous success rate 52%. Furthermore, planner 37%. algorithm 27%. Finally, average kilometers traveled before collision object increased 1.5 times. The main source code can be reached web page: https://heshameraqi.github.io/dynamic_cil_autonomous_driving .