t5lab.open_run¶
Open-terrain train+eval driver. Env vars: GRID(32), ITERS(300), NAG(10), SEED(0), CTRL(mvprop|greedy). Trains, then evals COVERAGE + OVERLAP + CONNECTIVITY (lambda2 / reach / connected-fraction) for greedy and sampled goal-selection. Writes a results JSON for the report. CTRL=greedy drops MVProp (deterministic step-toward-goal) — the no-planner control.