Territory Map – Vision-based GVF

An interactive demo of General Value Functions (GVFs) learned from a 1D vision cone as a robot explores a procedurally generated territory map. The map has four cell types: empty (white), gray, orange, and overlap. Three GVFs are learned online — predicting discounted future encounters of each type along the robot's facing direction.

Controls: click the map, then use arrow keys to move. Press R to teleport randomly. Adjust γ with the slider or change the seed to get a new map.

Empty Gray Orange Overlap
Loading…
Vision GVF — Facing ↑ | γ = 0.960
0.96