• Type:#article
  • Year read:#read2022
  • Subject:
  • Bibtex: @wurman2022
  • Bibliography: Wurman, P. R., Barrett, S., Kawamoto, K., MacGlashan, J., Subramanian, K., Walsh, T. J., Capobianco, R., Devlic, A., Eckert, F., Fuchs, F., Gilpin, L., Khandelwal, P., Kompella, V., Lin, H., MacAlpine, P., Oller, D., Seno, T., Sherstan, C., Thomure, M. D., … Kitano, H. (2022). Outracing champion Gran Turismo drivers with deep reinforcement learning. Nature, 602(7896), 223–228. https://doi.org/10.1038/s41586-021-04357-7

Example citation

Key takeaways

  • They trained an AI to not just race competitively, but to also follow the unofficial rules of sportsmanship.
  • Deep reinforcement learning
  • Two dimensions of control: accelerate/brake and steering left/right


This is an interesting contrast to If you only care about the binary outcome (win or loss) you dont have to win by a large margin like in Go, because here they wanted to place many cars near the top and also adhere to the game’s unofficial sportsmanship rules.