Deep Learning Architecture for a Wide Variety of Sensors

Autonomous Solutions, Inc. is seeking to develop a sensing architecture that fuses information from both deterministic and machine-learned algorithms to provide a sophisticated world model for autonomous vehicle road-following and obstacle detection. This architecture allows for safe and efficient navigation through road networks to accomplish a mission, both in military applications such as AGR and in civilian applications. With the recent technological advances in deep learning, it becomes necessary to integrate these state-of-the-art algorithms into a single framework allowing for intelligent use of their data. ASI seeks to accomplish this with an ego-centric probabilistic road-detection map allowing for redundancy not only between sensors but also between algorithms. The informaiton from multiple sensors will be fused to allow the road to be dtected along with sensing obstacles to allow the vehicle to accomplish the desired mission in a safe manner. Fusing LiDAR, camera, and radar data will allow the system to be robust to environments where a single sensor might fail.