01 · BUILD & PREDICTYour testing ground.
Collect the battery ϟ and bring it home ⌂. Design a challenge, then predict what the robot will learn.
Route connected
02 · OBSERVEWatch its decisions.
Try the untrained model. Then compare its choices after learning from thousands of varied examples.
Use One move under the map to step through a choice. Decision details explain the inputs and action preferences.
0 demonstrated moves
Arrow keys / WASD record examples. GPU learning starts in Step 3.
03 · TRAINTeach a useful skill.
Your drawn maze is an extra test. Training uses changing maps of this size.
- Short routes in open rooms
- Walls and changing goals
- Corridors and detours
Difficulty increases after a separate practice check. A final 100-maze exam stays outside training. Quick runs may stop before the harder levels.
Your previous model stays saved.
04 · TEST & COMPAREPut it to the test.
Train a model to see its measured results.
Each challenge generates a fresh maze. Interactive tests keep the model's weights frozen.
05 · EXPLAINWhat did it learn?
Compare familiar and unfamiliar maze results. What changed, and where does it still struggle?
Your report includes predictions, settings and measured results.