DeepTesla from MIT course corresponds to Udacity's Behavioral Cloning project from NVidia (Davenet-2) and MIT extends it by combining 3D convolutions with RNNs for better steering angle estimates; MIT's SegFuse corresponds to Semantic Segmentation with Berkeley's FCN at Udacity; Deep Reinforcement Learning for path planning and crash avoidance is unique at MIT, Udacity focuses on camera/LiDAR/radar as the main sensors whereas MIT mentions ultrasonic sensors for detecting condition of the road; MIT uniquely teaches how to sense state of driver from facial features detection; Udacity adds classical computer vision with HOG+SVM, tracking of objects with Extended/Unscented Kalman filters, driving using PID and MPC controllers, path planning using polynomial path approximations. They both mention some advanced object detection algorithms like R-CNN or SSD.

TL;DR: Once you finished Udacity, MIT gives you more wonderful topics you'd understand instantly (except for Deep Reinforcement Learning where you need to do some graduate-level coursework yourself).