Agile Autonomous Driving using End-to-End Deep Imitation Learning¶
Authors: Yunpeng Pan, Ching-An Cheng, Kamil Saigol, Keuntaek Lee, Xinyan Yan, Evangelos Theodorou, Byron Boots
Published: 2017 (Conference Paper)
Source: Robotics: Science and Systems (RSS)
arXiv: 1709.07174
DOI: 10.15607/RSS.2018.XIV.056
Summary¶
Uses a "traditional" autonomy stack (trajectory optimization with learned dynamics, Kalman filter state estimation, and an handcrafted cost function) as the expert policy, then trains a neural network to imitate it end-to-end, from pixels to torques. Demonstrates that a full autonomy stack can be "compressed into" or "represented by" a single neural network. Notably, the trained neural network can be deployed with less expensive compute hardware and a lower fidelty sensor suite than the original autonomy stack.
Abstract¶
We present an end-to-end imitation learning system for agile, off-road autonomous driving using only low-cost on-board sensors. By imitating a model predictive controller equipped with advanced sensors, we train a deep neural network control policy to map raw, high-dimensional observations to continuous steering and throttle commands. Compared with recent approaches to similar tasks, our method requires neither state estimation nor on-the-fly planning to navigate the vehicle. Our approach relies on, and experimentally validates, recent imitation learning theory. Empirically, we show that policies trained with online imitation learning overcome well-known challenges related to covariate shift and generalize better than policies trained with batch imitation learning. Built on these insights, our autonomous driving system demonstrates successful high-speed off-road driving, matching the state-of-the-art performance.
Links¶
Primary
Standard
Alternate
Tags¶
-
Imitation learning
-
End-to-end learning
-
Reinforcement learning
-
Deep learning
-
End-to-end
-
Trajectory optimization
-
Autonomous driving
-
DDP
-
Model-based