Control Autonomous Vehicle from Pixels
Built a convolutional neural network steering controller that maps raw RGB front-camera images directly to steering commands, bypassing intermediate perception and planning modules in an end-to-end fashion. Trained the policy via imitation learning on expert demonstrations, then iteratively improved it with DAgger by querying the expert on states visited by the learner to correct compounding errors from distribution shift. Evaluated in the CARLA simulator, achieving stable lane-following and turn-taking on routes unseen during training.
Dixant Mittal
My research interests include reinforcement learning, planning & search, large language models, and decision-making under uncertainty.