Learning to Drive (L2D) as a Low-Cost Benchmark for Real-World Reinforcement Learning

被引:2
|
作者
Viitala, Art [1 ]
Boney, Rinu [1 ]
Zhao, Yi [1 ]
Ilin, Alexander [1 ]
Kannala, Juho [1 ]
机构
[1] Aalto Univ, Dept Comp Sci, Espoo, Finland
关键词
D O I
10.1109/ICAR53236.2021.9659342
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
We present Learning to Drive (L2D), a low-cost benchmark for real-world reinforcement learning (RL). L2D involves a simple and reproducible experimental setup where an RL agent has to learn to drive a Donkey car around three miniature tracks, given only monocular image observations and speed of the car. The agent has to learn to drive from disengagements, which occurs when it drives off the track. We present and open-source our training pipeline, which makes it straightforward to apply any existing RL algorithm to the task of autonomous driving with a Donkey car. We test imitation learning, state-of-the-art model-free, and model-based algorithms on the proposed L2D benchmark. Our results show that existing RL algorithms can learn to drive the car from scratch in less than five minutes of interaction. We demonstrate that RL algorithms can learn from sparse and noisy disengagement to drive even faster than imitation learning and a human operator.
引用
收藏
页码:275 / 281
页数:7
相关论文
共 50 条
  • [1] ContainerGym: A Real-World Reinforcement Learning Benchmark for Resource Allocation
    Pendyala, Abhijeet
    Dettmer, Justin
    Glasmachers, Tobias
    Atamna, Asma
    [J]. MACHINE LEARNING, OPTIMIZATION, AND DATA SCIENCE, LOD 2023, PT I, 2024, 14505 : 78 - 92
  • [2] NeoRL: A Near Real-World Benchmark for Offline Reinforcement Learning
    Qin, Rong-Jun
    Zhang, Xingyuan
    Gao, Songyi
    Chen, Xiong-Hui
    Li, Zewen
    Zhang, Weinan
    Yu, Yang
    [J]. ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35, NEURIPS 2022, 2022,
  • [3] Real-world humanoid locomotion with reinforcement learning
    Radosavovic, Ilija
    Xiao, Tete
    Zhang, Bike
    Darrell, Trevor
    Malik, Jitendra
    Sreenath, Koushil
    [J]. SCIENCE ROBOTICS, 2024, 9 (89)
  • [4] Offline Learning of Counterfactual Predictions for Real-World Robotic Reinforcement Learning
    Jin, Jun
    Graves, Daniel
    Haigh, Cameron
    Luo, Jun
    Jagersand, Martin
    [J]. 2022 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA 2022), 2022, : 3616 - 3623
  • [5] Reinforcement Learning in Robotics: Applications and Real-World Challenges
    Kormushev, Petar
    Calinon, Sylvain
    Caldwell, Darwin G.
    [J]. ROBOTICS, 2013, 2 (03): : 122 - 148
  • [6] Real-World Reinforcement Learning via Multifidelity Simulators
    Cutler, Mark
    Walsh, Thomas J.
    How, Jonathan P.
    [J]. IEEE TRANSACTIONS ON ROBOTICS, 2015, 31 (03) : 655 - 671
  • [7] Learning real-world heterogeneous noise models with a benchmark dataset
    Sun, Lu
    Lin, Jie
    Dong, Weisheng
    Li, Xin
    Wu, Jinjian
    Shi, Guangming
    [J]. PATTERN RECOGNITION, 2024, 156
  • [8] REPLAB: A Reproducible Low-Cost Arm Benchmark for Robotic Learning
    Yang, Brian
    Jayaraman, Dinesh
    Zhang, Jesse
    Levine, Sergey
    [J]. 2019 INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA), 2019, : 8691 - 8697
  • [9] Real-world Robot Reaching Skill Learning Based on Deep Reinforcement Learning
    Liu, Naijun
    Lu, Tao
    Cai, Yinghao
    Wang, Rui
    Wang, Shuo
    [J]. PROCEEDINGS OF THE 32ND 2020 CHINESE CONTROL AND DECISION CONFERENCE (CCDC 2020), 2020, : 4780 - 4784
  • [10] Setting up a Reinforcement Learning Task with a Real-World Robot
    Mahmood, A. Rupam
    Korenkevych, Dmytro
    Komer, Brent J.
    Bergstra, James
    [J]. 2018 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS), 2018, : 4635 - 4640