Skip to content

Latest commit

 

History

History
85 lines (71 loc) · 2.82 KB

README.md

File metadata and controls

85 lines (71 loc) · 2.82 KB

Driver

The driver is made up of two components:

  • torcs_client is the compatibility client that allows for TORCS abstraction.
  • agents is the folder designated to contain the autonomous agents

pytorcs structure

Action space

The action space is variable in size, depending on how the environment is configured.

  • If throttle == False The action space is a 1-dimensional array, ranging from [-1,1]. The single value represents the steering request (-1 complete left, 1 complete right).

    action[0] := steering

  • If throttle == True The action space is a 2-dimensional array. The first value remains the steering.

    The second value is combined brake/throttle request:

    • [-1,0] for the brake
    • [0,1] for the throttle.
    action[0] := steering
action[1] := breaking/throttle

Sensor description

  • track - Vector of 19 range finder sensors: each sensors returns the distance between the track edge and the car within a range of 200 meters.
  • angle - Angle between the car direction and the direction of the track axis.
  • speedX - Speed of the car along its longitudinal axis.
  • speedY - Speed of the car along its transverse axis.
  • speedZ - Speed of the car along its Z axis.
  • wheelSpinVel - Vector of 4 sensors representing the rotation speed of wheels.
  • rpm - Number of rotation per minute of the car engine (could be used by the gearshift).
  • trackPos - Distance between the car and the track axis.
  • img - Vision from the driver perspective. By default a rgb 640x480 image.

Many other values are sent by torcs, but most of them are (currently) not useful for our purpose. The full list is:

angle
curLapTim
damage
distFromStart
distRaced
fuel
gear
lastLapTime
opponents
racePos
rpm
speedX
speedY
speedZ
track
trackPos
wheelSpinVel
z

More info on the sensors can be found in the original scr paper.

State space

pytorcs structure

The state space is variable too. Its size and composition can be changed in the simulation.yaml configuration file.

The returned observation at each step is a dictionary made up of numpy arrays of (normalized) sensor values, with the sensor name as key.

For example an observation may look like this:

observation["track"].shape = [19]
observation["angle"].shape = [1]
observation["speedX"].shape = [1]
observation["speedY"].shape = [1]
observation["speedZ"].shape = [1]
observation["wheelSpinVel"].shape = [4]
observation["rpm"].shape = [1]
observation["trackPos"].shape = [1]
observation["img"].shape = [640, 480, 3]

Customization

To use the environment with your custom run file or use it standalone follow this guide.