Simulation-based reinforcement learning for real-world autonomous driving

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Osiński, Błażej, Jakubowski, Adam, Miłoś, Piotr, Zięcina, Paweł, Galias, Christopher, Homoceanu, Silviu, Michalewski, Henryk
Format: Preprint
Published: 2019
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910396526362624
author Osiński, Błażej
Jakubowski, Adam
Miłoś, Piotr
Zięcina, Paweł
Galias, Christopher
Homoceanu, Silviu
Michalewski, Henryk
author_facet Osiński, Błażej
Jakubowski, Adam
Miłoś, Piotr
Zięcina, Paweł
Galias, Christopher
Homoceanu, Silviu
Michalewski, Henryk
contents We use reinforcement learning in simulation to obtain a driving system controlling a full-size real-world vehicle. The driving policy takes RGB images from a single camera and their semantic segmentation as input. We use mostly synthetic data, with labelled real-world data appearing only in the training of the segmentation network. Using reinforcement learning in simulation and synthetic data is motivated by lowering costs and engineering effort. In real-world experiments we confirm that we achieved successful sim-to-real policy transfer. Based on the extensive evaluation, we analyze how design decisions about perception, control, and training impact the real-world performance.
format Preprint
id arxiv_https___arxiv_org_abs_1911_12905
institution arXiv
publishDate 2019
record_format arxiv
spellingShingle Simulation-based reinforcement learning for real-world autonomous driving
Osiński, Błażej
Jakubowski, Adam
Miłoś, Piotr
Zięcina, Paweł
Galias, Christopher
Homoceanu, Silviu
Michalewski, Henryk
Machine Learning
Artificial Intelligence
Robotics
We use reinforcement learning in simulation to obtain a driving system controlling a full-size real-world vehicle. The driving policy takes RGB images from a single camera and their semantic segmentation as input. We use mostly synthetic data, with labelled real-world data appearing only in the training of the segmentation network. Using reinforcement learning in simulation and synthetic data is motivated by lowering costs and engineering effort. In real-world experiments we confirm that we achieved successful sim-to-real policy transfer. Based on the extensive evaluation, we analyze how design decisions about perception, control, and training impact the real-world performance.
title Simulation-based reinforcement learning for real-world autonomous driving
topic Machine Learning
Artificial Intelligence
Robotics
url https://arxiv.org/abs/1911.12905