Probabilistic and Reinforcement Learning Track
2023 International Planning Competition
2023 International Planning Competition
Race Car Control
| Example name | RaceCar |
| Action space | Dict |
| State space | Dict |
A race car is required to reach a target from an initial position. The race car is described as a kinematic 2nd order model. The race track is bounded with lines representing the track bounds, hitting one of the bounds is considered a failure and the episode is terminated. The shape of the track is represented by a series of point conected with lines.
| Constant | Type | Desc |
|---|---|---|
| X1(b) | float32 | boundary is the line segment (X1, Y1) -> (X2, Y2) |
| Y1(b) | float32 | boundary is the line segment (X1, Y1) -> (X2, Y2) |
| X2(b) | float32 | boundary is the line segment (X1, Y1) -> (X2, Y2) |
| Y2(b) | float32 | boundary is the line segment (X1, Y1) -> (X2, Y2) |
| X0 | float32 | starting x position of car |
| Y0 | float32 | starting y position of car |
| GX | float32 | x center of goal region |
| GY | float32 | y center of goal region |
| RADIUS | float32 | radius of goal region |
| COST | float32 | cost of fuel, proportional to force |
| GOAL_REWARD | float32 | reward upon reaching the goal region |
| MAX-F | float32 | maximum force in each direction |
| MASS | float32 | mass of the car |
| DT | float32 | rhow much time passes between epochs |
All of these can be read from the RDDLEnv interface and from the RDDL files.
The actioins are the force (or acceleartion in a kinematic setting) applied to each movement axis of the race car.
| Action | Type | Desc |
|---|---|---|
| fx | BOX(1, -MAX-F, MAX-F, float32) | x force component applied to the car |
| fy | BOX(1, -MAX-F, MAX-F, float32) | y force component applied to the car |
The state space is the positions and velocities of the race car in both axis.
| State | Type | Desc |
|---|---|---|
| x | Box(1, -inf, inf, float32) | x position of car |
| y | Box(1, -inf, inf, float32) | y position of car |
| vx | Box(1, inf, inf, float32) | c velocity of car |
| vy | Box(1, -inf, inf, float32) | y velocity of car |
The reward of this domain is the minus the cost of using the engines (min power cost) plus the goal reward if the agent reaches the goal