Collective Tree Exploration

This illustration shows several agents exploring an unknown tree, represented here as a 2D maze. All agents start from the upper-left cell and progressively claim cells as they discover them. The controls let you update maze shapes, team sizes, and exploration strategies. For more theoretical background, see this paper.

Maze generation

Wilson : uniform spanning tree, D ~ N.
Recursive DFS : longer paths, D ~ N^(5/4).
Comb : difficult for CTE.
Binary Tree : biased carving (north/east), diagonal flows.

Algorithms

Random DFS independent DFSs with random tie-breaking. No-communication.
CTE Greedy algorithm. Agents go to nearest unfinished subtree. Accelerated version (using complete communications).
LGA Locally greedy v3: reanchor with shared anchor counts.
OPT 2-approximation of offline algorithm. Full knowledge.

Suggested parameters

Try algorithm LGA with large number of robots (up to 50), max speeds, large N (up to 300), and shading unvisited nodes (see debug panel) to have nice (fractal) patterns emerge.

Elapsed time & Progress
0.00 s 0 moves 0 %
All speeds
20 / s
Debug
Performance
0 fps 0.0 ms/frame 0 % lag