End-to-End Autonomous Driving With Semantic Depth Cloud Mapping and Multi-Agent

Focusing on the task of point-to-point navigation for an autonomous driving vehicle, we propose a novel deep learning model trained with end-to-end and multi-task manners to perform both perception control tasks simultaneously. The is used drive ego vehicle safely by following sequence routes defined global planner. part encode high-dimensional observation data provided RGBD camera while performing semantic segmentation, depth cloud (SDC) mapping, traffic light state stop sign prediction. Then, decodes encoded features along additional information GPS speedometer predict waypoints that come latent feature space. Furthermore, two agents are employed process these outputs make policy determines level steering, throttle, brake as final action. evaluated CARLA simulator various scenarios made normal-adversarial situations different weathers mimic real-world conditions. In addition, do comparative study some recent models justify performance in multiple aspects driving. Moreover, also conduct ablation SDC mapping multi-agent understand their roles behavior. As result, our achieves highest score even fewer parameters computation load. To support future studies, share codes at https://github.com/oskarnatan/end-to-end-driving .