The Truck Backer-Upper
This lecture covers the concepts of emulation of kinematics from observations and training a policy. It is a part of the Deep Learning Course at CDS, a course that covered the latest techniques in deep learning and representation learning, focusing on supervised and unsupervised deep learning, embedding methods, metric learning, convolutional and recurrent nets, with applications to computer vision, natural language understanding, and speech recognition. Prerequisites for this module include: Models 1-6 of this course and Introduction to Data Science or a Graduate Level Machine Learning.
Chapters:
00:00 – Welcome to class
00:00:28 – Action plan
00:05:05 – State transition equations (recap)
00:07:13 – The Truck Backer-Upper
00:09:16 – Vehicle configuration
00:11:47 – Implementation in a Jupyter Notebook
00:13:52 – Manual parking tests
00:18:03 – Training: a two-stage learning process
00:20:57 – State update equations for a trailer truck
00:22:59 – Emulator training strategy
00:26:27 – Training protocol (I)
00:31:14 – Control as RNN (again)
00:35:37 – Training protocol (II)
00:38:16 – Unrolling in time (AKA BPTT)
00:39:59 – Successful controller's trajectories
00:42:03 – Additional resources
00:42:56 – PyTorch (partial) implementation
00:50:32 – Bayesian neural nets
00:53:34 – Dropout
00:55:59 – Uncertainty for a regressor (demo)
01:00:35 – And that was it :D
