Explore the intricacies of the Truck Backer-Upper problem in this comprehensive lecture by Alfredo Canziani. Delve into state transition equations, vehicle configurations, and implementation in Jupyter Notebook. Learn about the two-stage learning process for training, including emulator training strategies and control as RNN. Discover unrolling in time (BPTT) techniques and examine successful controller trajectories. Gain insights into PyTorch implementation, Bayesian neural networks, dropout, and uncertainty in regression models. This in-depth session covers both theoretical concepts and practical demonstrations, providing a thorough understanding of advanced machine learning applications in vehicle control systems.
Overview
Syllabus
– Welcome to class
– Action plan
– State transition equations recap
– The Truck Backer-Upper
– Vehicle configuration
– Implementation in a Jupyter Notebook
– Manual parking tests
– Training: a two-stage learning process
– State update equations for a trailer truck
– Emulator training strategy
– Training protocol I
– Control as RNN again
– Training protocol II
– Unrolling in time AKA BPTT
– Successful controller's trajectories
– Additional resources
– PyTorch partial implementation
– Bayesian neural nets
– Dropout
– Uncertainty for a regressor demo
– And that was it :D
Taught by
Alfredo Canziani
