t5lab.render_planner¶
Visualize a distilled planner: render its value field (heat) with an agent descending it from a far start toward the goal — MVProp reaches; MSP gets trapped in a local maximum. The heatmap IS the learned field (bright = high value = near goal), so the viz shows both the mechanism and the failure.
PYTHONPATH=.:../../../FiedlerValueEstimation ARCH=mvprop MAP=rooms SEED=3 OUT=gifs/plan_mvprop.gif $PY -m t5lab.render_planner