File size: 1,326 Bytes
929e74d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
/home/cercola/matte_server/DiffusionControl/nonlinear/data_generation/pendulum_lmi.py:65: UserWarning: Solution may be inaccurate. Try another solver, adjusting the solver settings, or solve with verbose=True for more information.
  prob.solve(solver=cp.SCS, verbose=False)

=== CONSTRAINED OC ===
Constrained optimal control on the pendulum (held-out N=10, cap θ≤0.24)

PRIOR (quadratic-trained, constraint-unaware):
  satisfies |θ|≤0.24: 0.30   median peak |θ| = 0.272   median effort 18.557
STEERED toward  min ∫u²  s.t. |θ|≤0.24  (barrier λ=10000.0):
  satisfies: 0.30   median peak |θ| = 0.264   median effort 18.827   certified 1.00
  effort REGRET vs per-plant constrained oracle: median +0.5% (n=3)

effort-vs-cap constrained-optimal frontier (steered):
  θ_cap   mean effort (satisfying)   satisfied
  0.20    nan                    0.00
  0.24    7.831                    0.30
  0.28    21.301                    1.00
  0.32    21.227                    1.00

INTERPRETATION: a non-quadratic, hard-state-constrained optimal-control problem solved
through the differentiable non-linear pendulum: the constraint-unaware prior violates the
cap; steering enforces it at minimum effort (near the per-plant constrained optimum),
certified; tighter caps cost more effort (the constrained-optimal frontier).