- 최신
- 최다 투표
- 가장 많은 댓글
Hi,
You can get some samples to improve speed here:
- https://blog.gofynd.com/how-we-broke-into-the-top-1-of-the-aws-deepracer-virtual-circuit-573ba46c275
- https://medium.com/analytics-vidhya/training-deepracer-for-speed-52007aa03af5
- https://dominguezdaniel.medium.com/aws-deepracer-parameters-of-reward-functions-4b6f4b7e9eaf
- https://refactored.ai/microcourse/notebook?path=content%2FDeepRacer%2FAWS_DeepRacer_Reward_function_Additional_material.ipynb
- https://www.linkedin.com/pulse/samples-reward-functions-aws-deepracer-bahman-javadi/
Best,
Didier
Please specify if you are in the deepracer league or deepracer student league. It would fundementally change our answer to you. What a lot of people do not grasp is speed is a fixed value from your action space. People will see the video and think they can reward above max speed. The other issue is reward = speed or reward = speed + some other reward behavior are bad reward function.
Without going into the math these reward function will prompt the car to go slower because highest overall reward is achieved. Ultimately the cars goal is to get the highest reward. This means analyzing your reward function is vital to understand if you shaped your reward correctly. Lower steps will be faster laps.
Thank you for your response. I am a participant in the Deepracer student league. Currently, I am utilizing the relative speed and reward features in one of my models, also another one with the step functionality with a maximum of 300 steps. However, I am curious to know if there are alternative options for achieving maximum speed if we do not use the speed parameter in this way. Can you suggest any other approaches?
Thank you for providing resources, they were very helpful to me.