Find Treasure
The agent is tasked to navigate through a partially observable maze to retrieve a treasure. The agents receive a +10 reward upon reaching the treasure, and a constant -1 time penalty for each step taken. The observation space is a top-down accumulated view of where the agent has explored, initialized by a square area around the agent. The action space is a two-dimensional vector of the next destination coordinates, to which a low-level planning algorithm will navigate the agent. The max episode length is 15 seconds.
Action Space
A 2-dimensional continuous action space:
NEXT_DESTINATION_X(0)NEXT_DESTINATION_Y(1)
Observation Space
320x320 camera view with bounding box:
Unity Parameters
Game specific parameters:
| Parameter Name | Parameter Format | Description |
|---|---|---|
-NumAgents |
integer | Number of AI agents |
-EventChannelID |
string | Event side channel assigned by Algorithms |
-GameStopChannelID |
string | GameStop side channel assigned by Algorithms |
Universal parameters:
| Parameter Name | Parameter Format | Description |
|---|---|---|
-NakamaID |
string | Server ID assigned by Instance Server |
-DojoScreenSize |
intxint |
Set window size on launch |
-DojoScreenPos |
int,int |
Set window position on launch |
-DojoMonitorID |
integer | Set window monitor on launch |
-CaptureSizeW |
integer | Camera capture size (width) |
-CaptureSizeH |
integer | Camera capture size (height) |
-DojoRecording |
none | Flag to enable recording |
-DojoRecordingFile |
string | To overwrite recording log file path |
-DojoRecordingIdentity |
string | To overwrite recording log identity |
User interface controls:
| Action Name | Keyboard Mapping |
|---|---|
| Toggle UI | ESC |