NVIDIA's Autonomous Vehicle Research Lab will present breakthroughs in developing and testing deep neural networks to improve the safety and robustness of self-driving cars. One approach involves teaching a deep convolutional neural network (DNN) to drive by observing human drivers and emulating their behavior for lane keeping, lane changes, and turns. In addition, the session will showcase tools used to visualize the data processing of the neural network during training and testing, as well as the use of simulation to enhance the training process. This technology is part of an end-to-end platform that will ultimately enable self-driving cars up to Level 5. Finally, the session will cover how DNNs can learn autonomous driving related tasks that were previously thought solvable only by manual decomposition of the problem, and how learned execution of maneuvers can be performed without relying solely on localization and HD maps.
Hall: Hall D