Maze Task Specifications
[Home]
Tasks are designed to test the cognitive capabilities of a virtual mouse in certain terrains.
The tasks are set up according to:
- Cues for self-location
- Complexity of the terrain
- Gradual rise of difficulty in the same terrain
Checkered walls and landmarks are placed as cues.
A linear maze, a cross maze, and an arrow-head maze are used as terrains.
In the cross maze, the start position is variable so that the mouse is not to follow the determined path.
The arrow-head maze is designed so that it cannot be mapped with regular triangles, for the grid cells in the brain represent regular triangles.
In the linear maze, task difficulty varies.
M Sato et al., eNeuro (2017) http://www.eneuro.org/content/4/3/ENEURO.0369-16.2017
Video of a mouse VR experiment: https://www.youtube.com/watch?v=uEM0a5utXGo
- The mouse agent moves at 0.15 sec/step {1m/step or ±10 deg./step}.
- S is the start point for the episode.
- G is the goal point.
- The next episode begins if a certain condition is met.
- After Task 2, if a certain condition is met at R, reward can be obtained at R and G.
- Else a certain condition is met at R, reward cannot be obtained at G either.
- Maze length: 24m
- S: 6, 4, 2, 4, 2, 4, 2 :G (m)
- S: White, Green, White, Green, White, Green, White: G
- Maze width: 1m
- Wall color
- Green wall: Green-black checkered wall (in 3 locations between S and G)
- White wall: White-black checkered wall
- The landmarks (Mountain、Tower、Building) beyond the wall are the same as those in the video (Fig. Right).
| No. | Tasks | Aims |
|---|---|---|
| 1 | Reward is unconditionally given at G. | To learn to get reward by reaching G |
| 2 | Reward is given when passing R and afterward at G. | To learn to get reward by reaching R besides G |
| 3 | Reward is given at R having waiting for two seconds there and afterward at G. Reward is not given if it does not wait at R (either at R nor G). | To learn to get reward after waiting for 2 seconds at R |
| 4 | R moves to another location. R moves to the green wall closest to G. Reward condition is the same as 3. | To learn to get reward after waiting for 2 seconds at R even if its location changes |
| 5 | R moves to the green wall closest to S. Reward condition is the same as 4. | To learn to get reward after waiting for 2 seconds at R even if its location changes further |
| 6 | The location where an interim reward is given changes each time. | To generalize the tasks 4 and 5 |
| 7 | Geometrical figures (○、△、□) are shown to the mouse agent at S. ○: Reward is given if it waits for 2 seconds at the green wall closest to G. △: Reward is given if it waits for 2 seconds at the green wall in the middle. □: Reward is given if it waits for 2 seconds at the green wall closest to S. | To learn the relation between the figures and locations; the condition for reward is the same. |
| 8.a-d | All wall parts become white-black checkered. | To learn the relation between the location (not the color) and reward |
| 8.a | If it waits for 2 seconds at a location that was a green wall in the middle, reward is given and afterwards at G (same as 4) | - |
| 8.b | The location where reward is given changes as in 5. | - |
| 8.c | The location where reward is given changes as in 6. | - |
| 8.d | Geometrical figures are shown at S as in 7. | - |
- Maze arm length (from the center to the end): 12m
- Maze width: 1m
- Tower in the north-east
- Mountain in the south-west
| No. | Tasks | Aims |
|---|---|---|
| 1 | S is at the end of the arm. An R is fixed. There are three Ss and the start location is randomly chosen. | To grasp reward location allocentrically rather than egocentrically |
| 2 | S is at the center of the cross. Rs locate at the ends of the arm. All rewards (3 of them) should be obtained. | To learn that there is no reward at the location already visited |
Task 1
Task 2
- S is the start location.
- Reward is given at R having waiting for two seconds there R.
- An episode ends having reached G after meeting the reward condition at R.
- The episode is terminated unless the condition is not met after a certain number of steps.
- The middle path cannot be passed at the beginning.
- The bottom route is shorter.
- The central bottom angle of the triangle: 120 degs.
| No. | Tasks | Aims |
|---|---|---|
| 1 | No middle path so that G can be reached from above or below. | To learn to reach G via the shortest path while obtaining reward with the same condition as the linear maze |
| 2 | A red wall appears and G becomes unreachable from below. | To learn to reach G via another path even the shortest path is blocked while obtaining reward with the same condition as the linear maze |
| 3 | The middle path is formed. | To learn to reach G via the middle path as the new shortest path while obtaining reward with the same condition as the linear maze |
Task 1
Task 2
Task 3
