Location Task

Calculates a sparse distance reward based on the distance to a location or an actor. Can maximize or minimize a distance. A reward of 1 is only given only if the agent reaches the goal target within the goal distance.

Configuration

Each of the following parameters can be placed in the configuration field for a location task sensor (see scenario files.)

LocationActor

The reward is given based on the distance between this actor and the goal target. Defaults to the task’s agent.

"LocationActor": "name-of-actor"

GoalTarget

The Location task needs an actor or a location to use in the distance calculations.

"GoalActor": "name-of-actor"

or

"GoalLocation": [3.14, 2.71, 117]

GoalDistance

This is the distance from the goal target the LocationActor must be to get a reward and terminal.

"GoalDistance": 1024.0

NegativeReward

A boolean representing whether reaching the goal target returns 1 or -1. Defaults to false.

"NegativeReward": false

HasTerminal

A boolean representing whether reaching the goal target returns a terminal value or not. Defaults to false.

"HasTerminal": true

Example

{
    "LocationActor": "golf-ball",
    "GoalActor": "cup",
    "GoalDistance": 1024.0,
    "NegativeReward": false,
    "HasTerminal": true
}