Data Gatherer for Reinforcement Learning

A plugin by Palamabron

Data Gatherer for Reinforcement Learning
Downloads 187
Updated 5 months ago
Released 9 months ago
Created 9 months ago
Numeric ID 421
Min. Openplanet 1.26.0
Signed
Game Trackmania

Plugin sends to data through a socket (using port 9000) information:

  1. speed
  2. map progress (only for tmrl-test now)
  3. position x, y and z of a vehicle
  4. input steer
  5. input gas pedal
  6. input brake
  7. race state (isFinished)
  8. acceleration
  9. jerk
  10. engine rpm
  11. aim yaw
  12. aim pitch
  13. FL and FR SteerAngle
  14. FL and FR WheelRot
  15. FL, FR, RL and RR DamperLen
  16. FL, FR, RL and RR SlipCoef
  17. ReactorAirControl x, y and z
  18. GroundDist
  19. IsCrashed
  20. IsReactorGroundMode
  21. IsGroundContact
  22. Engine Current Gear
  23. FL, FR, RL and RR GroundContactMaterial

Can be used in reinforcement learning (plugin based on tmrl) My version of tmrl (including TQC algorithm here)

Changelog

2.2.0 - 5 months ago

Changed speed for relative value.

2.1.1 - 7 months ago

Changing inputs for better model training.

2.1.0 - 8 months ago

Changed attributes to:

  1. speed
  2. distance
  3. position (x, y, z)
  4. input_steer
  5. input gas pedal
  6. input brake
  7. acceleration
  8. jerk
  9. rpm
  10. aim yaw
  11. aim pitch
  12. steer angle
  13. slip coefficient (front left and right wheel)
  14. crashed (boolean, based on jerk value)
  15. current gear
  16. current checkpoint number
  17. current lap Every value is converted to float. Some of the previous data were periodic or highly correlated with other attributes, which made training a neural network much harder.

View all changelogs