A Champion-level Vision-based Reinforcement Learning Agent for Competitive Racing in Gran Turismo 7
Authors:
- Hojoon LeeHLHojoon Lee
- Takuma SenoTSTakuma Seno
- Jun Jet TaiJTJun Jet Tai
- Kaushik SubramanianKSKaushik Subramanian
- Kenta KawamotoKKKenta Kawamoto
- Peter StonePSPeter Stone
- Peter R. WurmanPWPeter R. Wurman
Preprint
Abstract
Deep reinforcement learning has achieved superhuman racing performance in high-fidelity simulators like Gran Turismo 7 (GT7). It typically utilizes global features that require instrumentation external to a car, such as precise localization of agents and opponents, limiting real-world applicability. To address this limitation, we introduce a vision-based autonomous racing agent that relies solely on ego-centric camera views and onboard sensor data, eliminating the need for precise localization during inference. This agent employs an asymmetric actor-critic framework: the actor uses a recurrent neural network with the sensor data local to the car to retain track layouts and opponent positions, while the critic accesses the global features during training. Evaluated in GT7, our agent consistently outperforms GT7's built-drivers. To our knowledge, this work presents the first vision-based autonomous racing agent to demonstrate champion-level performance in competitive racing scenarios.Keywords
All Related Versions
This publication has 18 references indexed in Scilit:
- AnyCar to Anywhere: Learning Universal Dynamics Model for Agile and Adaptive Mobility2025
- Autonomous Vehicles on the Edge: A Survey on Autonomous Vehicle RacingIEEE Open Journal of Intelligent Transportation Systems, 2022
- A survey of state-of-the-art on visual SLAMExpert Systems with Applications, 2022
- Scenario Understanding and Motion Prediction for Autonomous Vehicles—Review and ComparisonIEEE Transactions on Intelligent Transportation Systems, 2022
- Outracing champion Gran Turismo drivers with deep reinforcement learningNature, 2022
- Learn-to-Race: A Multimodal Control Environment for Autonomous Racing2021
- Vision-Based Autonomous Car Racing Using Deep Imitative Reinforcement LearningIEEE Robotics and Automation Letters, 2021
- Vehicle Odometry with Camera-Lidar-IMU Information Fusion and Factor-Graph OptimizationJournal of Intelligent & Robotic Systems, 2021
- Super-Human Performance in Gran Turismo Sport Using Deep Reinforcement LearningIEEE Robotics and Automation Letters, 2021
- MuJoCo: A physics engine for model-based control2012