PhD Defense: LEARNING-BASED AUTONOMOUS DRIVING WITH ENHANCED DATA EFFICIENCY AND POLICY TRAINING

Autonomous vehicles are capable of sensing their environment and moving safely with little to no human input. They will impact our means of transportation and ways of life in years to come. Increasingly autonomous driving is adopted in real-world applications, e.g. autonomous truck for cargo transportation, self-driving taxi in an urban area, etc. With the rapid advances in hardware and software design, learning-based autonomous driving is becoming a viable and popular solution. As commonly known, data is central to all learning-based methods. We aim to improve performance by utilizing self-augmented data (data augmentation and adversarial learning), other modalities' data (multi modality learning and auxiliary modality learning), and other domains' data (transfer learning and domain adaptation). In addition, while input data is the key component of autonomous driving in the front-end, policy is also an important component in the back-end, which actually controls the vehicle to navigate safely. We thus address the issue on policy learning with enhanced inverse reinforcement learning.

Examining Committee

Chair:

Dr. Ming Lin

Dean's Representative:

Dr. Mumu Xu

Members:

Dr. Tom Goldstein

Dr. Furong Huang

Dr. Dinesh Manocha

Dr. Tianyi Zhou