Download results/pendulum/hardness_full.log from matCercola18/DiffusionControl: direct link, hf CLI and curl.
- Browser
- Download file 3.2 kB
-
https://huggingface.co/matCercola18/DiffusionControl/resolve/main/results/pendulum/hardness_full.log
- Command line
-
hf download hf://matCercola18/DiffusionControl/results/pendulum/hardness_full.log
-
curl -L -o hardness_full.log https://huggingface.co/matCercola18/DiffusionControl/resolve/main/results/pendulum/hardness_full.log
3.2 kB
| pend 0 (l=1.14,m=0.33,b=0.18): feasible=True fail 2/24=8% stab-cost spread 95% modes~4 16.9s | |
| pend 1 (l=0.77,m=0.41,b=0.11): feasible=True fail 0/24=0% stab-cost spread 103% modes~3 16.3s | |
| pend 2 (l=0.54,m=0.90,b=0.19): feasible=True fail 10/24=42% stab-cost spread 320% modes~6 16.5s | |
| pend 3 (l=0.52,m=0.88,b=0.11): feasible=True fail 9/24=38% stab-cost spread 172% modes~4 16.4s | |
| pend 4 (l=1.31,m=0.85,b=0.13): feasible=True fail 18/24=75% stab-cost spread 350% modes~5 16.2s | |
| pend 5 (l=1.41,m=0.65,b=0.27): feasible=True fail 18/24=75% stab-cost spread 310% modes~5 16.3s | |
| pend 6 (l=1.11,m=1.20,b=0.08): feasible=True fail 18/24=75% stab-cost spread 64% modes~4 16.3s | |
| pend 7 (l=1.23,m=1.18,b=0.19): feasible=True fail 15/24=62% stab-cost spread 495% modes~6 16.1s | |
| pend 8 (l=1.04,m=0.92,b=0.04): feasible=True fail 15/24=62% stab-cost spread 255% modes~6 16.3s | |
| pend 9 (l=1.44,m=0.89,b=0.25): feasible=True fail 15/24=62% stab-cost spread 410% modes~8 16.2s | |
| pend 10 (l=1.32,m=0.92,b=0.24): feasible=True fail 16/24=67% stab-cost spread 139% modes~5 16.5s | |
| pend 11 (l=0.50,m=0.65,b=0.09): feasible=True fail 0/24=0% stab-cost spread 137% modes~6 17.7s | |
| pend 12 (l=1.36,m=0.42,b=0.27): feasible=True fail 10/24=42% stab-cost spread 157% modes~5 18.1s | |
| pend 13 (l=0.53,m=0.95,b=0.04): feasible=True fail 12/24=50% stab-cost spread 208% modes~6 18.2s | |
| pend 14 (l=1.23,m=0.77,b=0.11): feasible=True fail 18/24=75% stab-cost spread 19% modes~3 17.0s | |
| pend 15 (l=0.68,m=0.58,b=0.06): feasible=True fail 0/24=0% stab-cost spread 143% modes~4 17.4s | |
| pend 16 (l=1.36,m=0.74,b=0.15): feasible=True fail 14/24=58% stab-cost spread 438% modes~7 16.9s | |
| pend 17 (l=1.04,m=1.10,b=0.24): feasible=True fail 16/24=67% stab-cost spread 433% modes~5 16.7s | |
| pend 18 (l=0.80,m=1.14,b=0.08): feasible=True fail 13/24=54% stab-cost spread 300% modes~7 16.1s | |
| pend 19 (l=0.92,m=0.62,b=0.03): feasible=True fail 16/24=67% stab-cost spread 43% modes~3 16.1s | |
| === PENDULUM HARDNESS === | |
| inverted-pendulum family — per-instance multi-start synthesis (n_pend=20, k_inits=24, steps=100, θ0=0.25 rad) | |
| feasible (a tuned linear φ stabilises): 20/20 | |
| Among FEASIBLE pendulums: | |
| FAIL-to-stabilise rate across random inits: median 60% (mean 49%, max 75%) | |
| -> even on STABILISABLE plants, a random init lands in the DIVERGENT basin this often. | |
| cost spread among STABILISED inits: median 190% (max 495%) | |
| -> multiple local minima (>1 cost mode on 20/20 feasible pendulums). | |
| per-pendulum multi-start solve time: median 16.4s (24 inits × 100 steps). | |
| CONVEX CONTRAST — same multistart on LTI+sector loops (LMI-stabilisable): | |
| fail rate median 0% vs pendulum 60%; cost spread median 0% vs pendulum 190%. | |
| -> the LTI loop is reliably stabilised from random inits; the pendulum is not. | |
| INTERPRETATION: unlike LTI+sector (convex LMI), per-instance pendulum synthesis is | |
| non-convex — random single-shot synthesis fails often and multi-start is needed | |
| (slow/unreliable). An amortised generative prior supplies a good init / a distribution | |
| of stabilising controllers, which is exactly the value §6r's warm-start demonstrated. | |