SOS Backup Shield for Learned Policies / results.json
Mechanism confirmed, baseline not beaten
1{
2 "seed": 7,
3 "math_check": {
4 "P": [
5 [
6 1.2500000000000002,
7 0.25
8 ],
9 [
10 0.25000000000000006,
11 0.37500000000000006
12 ]
13 ],
14 "identity_error": 1.1102230246251565e-16,
15 "gamma": 0.002734749554788495,
16 "max_backup_u_on_certificate": 0.17404742623406258,
17 "observed_min_neg_vdot_over_V": 0.7596534626874615,
18 "predicted_rate_bound": 0.7596526541079154,
19 "certificate_saturation_free": true
20 },
21 "prediction_1_dissipation": "-Vdot/V >= 1/lambda_max(P) on unsaturated certificate",
22 "prediction_2_horizon": "backward reachable volume is nondecreasing with T",
23 "prediction_3_scaling": "quadratic certificate area scales as terminal_radius^2",
24 "certificate_scaling_sweep": [
25 {
26 "terminal_radius": 0.03,
27 "certified_area": 0.0033698514343115014,
28 "area_over_base": 0.25,
29 "predicted_ratio": 0.25
30 },
31 {
32 "terminal_radius": 0.06,
33 "certified_area": 0.013479405737246005,
34 "area_over_base": 1.0,
35 "predicted_ratio": 1.0
36 },
37 {
38 "terminal_radius": 0.12,
39 "certified_area": 0.05391762294898402,
40 "area_over_base": 4.0,
41 "predicted_ratio": 4.0
42 },
43 {
44 "terminal_radius": 0.18,
45 "certified_area": 0.12131465163521406,
46 "area_over_base": 9.0,
47 "predicted_ratio": 9.0
48 }
49 ],
50 "reachability_sweep": [
51 {
52 "T": 0.5,
53 "volume": 0.025200000000000007,
54 "fraction": 0.0018812147272238647
55 },
56 {
57 "T": 1.0,
58 "volume": 0.08280000000000001,
59 "fraction": 0.006181134103735555
60 },
61 {
62 "T": 2.0,
63 "volume": 0.6228000000000001,
64 "fraction": 0.046492878258532655
65 },
66 {
67 "T": 3.0,
68 "volume": 3.387600000000001,
69 "fraction": 0.2528890083310938
70 },
71 {
72 "T": 4.0,
73 "volume": 8.341200000000002,
74 "fraction": 0.6226820747110992
75 }
76 ],
77 "input_saturation_sweep": [
78 {
79 "kp": 0.5,
80 "saturated_fraction": 0.7165775401069518
81 },
82 {
83 "kp": 1.0,
84 "saturated_fraction": 0.7486631016042781
85 },
86 {
87 "kp": 2.0,
88 "saturated_fraction": 0.6737967914438503
89 },
90 {
91 "kp": 4.0,
92 "saturated_fraction": 0.9090909090909091
93 },
94 {
95 "kp": 8.0,
96 "saturated_fraction": 0.9090909090909091
97 }
98 ],
99 "certified_rollout_audit": {
100 "n": 500,
101 "safe_rate": 1.0,
102 "terminal_rate": 1.0,
103 "max_violation": 0.0,
104 "horizon": 4.0
105 },
106 "policy_comparison": {
107 "n": 1000,
108 "baseline_violation_rate": 0.962,
109 "shield_violation_rate": 0.228,
110 "baseline_mean_violation": 14.94174058032316,
111 "shield_mean_violation": 0.08638416745996397,
112 "shield_mean_switches": 628.568
113 }
114}