Vision
- The vision system improved over the three CABots but
followed the same basic architecture.
- Input was a binary bitmap conversion of the jpg captured from the
agent in the environment.
- The retina was a collection of on-off sensors.
- The retina projected to V1 which did line, angle, and edge
detection.
- V1 projected to object recognition.
- In CABot1 and 2 we did pyramids and stalactites (up and down
facing tetrahedrons).
- In CABot3 we added texture recognition with grid cells, and
managed to do corridors based on lighting.
- Objects were recognised position dependently and position
independently.
- Some additional work derived from this managed to learn 5
different 2D shapes.
