Эксперимент: Trajectory-Learned Actuator-Aware Funnel Network (#357)

{ "worked": true, "confidence": 8, "verdict": "Built a state-only quantile funnel estimator and actuator-aware scalar saturated-integrator controller MVP in funnel_mvp.py. The mechanism checks succeeded: observed gain clipped exactly at the predicted k_max=ubar/rho, authority scaling was linear, and the normalized-error derivative identity had 0.2% relative error. However, the funnel-aware controller did not beat the fixed-gain saturated baseline on this setup, so the mechanism was verified but no performance win was observed.", "metrics": { "baseline": "fixed_gain_saturated: mean final absolute error 0.1285, mean maximum normalized error 0.8900, mean violation fraction 0.3125, mean absolute control 0.1918", "idea": "funnel_aware: mean final absolute error 0.1363, mean maximum normalized error 0.9438, mean violation fraction 0.3447, mean absolute control 0.1879; gain boundary predicted/observed 0.3891/0.3891; authority-scaling relative error 0.0 across ubar 0.15-0.70; normalized-error identity relative error 0.00198" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 funnel_mvp.py", "files": [ "funnel_mvp.py", "results.json" ], "limitations": "This is a scalar toy MVP rather than a trained neural network, and it uses a known integrator instead of learned or differentiable dynamics. The inferred funnel shrinks over time, held-out initial states include points outside its initial radius, and no multidimensional, noisy-dynamics, action-imitation, or training-compute comparison was tested." }