Works on safe, efficient autonomy using human-in-the-loop reinforcement learning; neuromorphic computing with spiking neural networks and AI hardware-software co-design for low-power, real-time inference; and robust training methods including debiased semi-supervised learning.