HPC-Enabled Deep-Learning and Simulation for AV Development

DOE’s most powerful and innovative High Performance Computing ( HPC) resources will be applied to the challenge of developing efficient and safe perception and control algorithms for autonomous vehicle operation in a greatly expedited process. The core approach will be first to conduct deep-learning analysis of vehicle sensor data sets with new methods such as ORNL’s MENNDL, using high performance computing, to accelerate the discovery of efficient analysis and control algorithms. A second phase will develop a virtual environment to test the algorithms for millions of scenarios and miles in faster than real time simulations. This virtual environment will be developed with support of human- and hardware-in-the-loop systems. Additional benefits will be derived from adapting the analytics methods to prognostics of truck and car electronic systems (whether autonomous or not) for vehicle “self-awareness” regarding fuel efficiency and cyber security.