Lamapi commited on
Commit
573eba1
·
verified ·
1 Parent(s): 70fb343

Publish Next Balance

Browse files
Files changed (10) hide show
  1. .gitattributes +2 -0
  2. README.md +65 -0
  3. balance_core.py +120 -0
  4. evaluation.json +1309 -0
  5. inference.py +61 -0
  6. model.json +1 -0
  7. nlrt.py +226 -0
  8. preview.png +3 -0
  9. replay.mp4 +3 -0
  10. selftest.json +1 -0
.gitattributes CHANGED
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ preview.png filter=lfs diff=lfs merge=lfs -text
37
+ replay.mp4 filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - control
5
+ - adaptive-control
6
+ - simulation
7
+ - next-lab
8
+ ---
9
+
10
+ # Next Balance
11
+
12
+ Next Balance keeps a simulated lift platform at 0.5 m while its payload changes at t≈4 s and friction rises at t≈7 s.
13
+ It compares a **fixed PID** controller with an **adaptive controller** that estimates mass, viscous friction and
14
+ Coulomb friction online (recursive least squares on filtered signals) and rescales its control law. **Simulation only.**
15
+
16
+ **Space:** https://huggingface.co/spaces/thelamapi/Next-Balance — set load change, actuator delay, sensor noise and friction; both controllers re-run in the browser.
17
+
18
+ ## Results on held-out scenarios (200 seeds per split, same disturbances for both controllers)
19
+
20
+ | Split | Controller | Max deviation after load change, mean / p90 (m) | Settling time, median (s) | Settled | Control energy (kN²·s) |
21
+ |---|---|---:|---:|---:|---:|
22
+ | nominal | Fixed PID | 0.144 / 0.283 | 1.97 | 81% | 0.061 |
23
+ | nominal | Adaptive (Next Balance) | 0.140 / 0.263 | 5.03 | 96% | 0.052 |
24
+ | stress | Fixed PID | 0.286 / 0.540 | 2.19 | 68% | 0.098 |
25
+ | stress | Adaptive (Next Balance) | 0.292 / 0.504 | 5.10 | 90% | 0.094 |
26
+
27
+ Stress split by condition (mean max deviation in m, settled rate):
28
+
29
+ | Condition | n | PID | Adaptive | PID settled | Adaptive settled |
30
+ |---|---:|---:|---:|---:|---:|
31
+ | heavy_load_change(dm>4kg) | 121 | 0.322 | 0.406 | 73% | 88% |
32
+ | load_removed(dm<0) | 10 | 0.234 | 0.035 | 20% | 100% |
33
+ | long_delay(>=40ms) | 88 | 0.447 | 0.359 | 51% | 82% |
34
+ | high_noise(>2mm) | 143 | 0.294 | 0.310 | 65% | 90% |
35
+
36
+ Paired wins (adaptive better on the same scenario): nominal max deviation 86/200,
37
+ stress max deviation 69/200,
38
+ stress energy 148/200.
39
+ Settling time is measured from the load change until the platform last leaves a ±1 cm band (the later friction change counts).
40
+ Runs that leave the band within the final 0.5 s count as not settled.
41
+
42
+ ![Two held-out stress scenarios](preview.png)
43
+
44
+ [Replay video](replay.mp4): the first held-out stress scenario and the stress scenario with the adaptive controller's largest deviation.
45
+
46
+ ## Method
47
+ * Plant: m(z̈+g) = u(t−delay) − c·ż − F·tanh(ż/0.01), dt = 10 ms, force limited to [0, 400] N, platform 2.0 kg.
48
+ * Shared α-β estimator (α=0.3, β=0.05). Both controllers only know the nominal payload.
49
+ * Fixed PID: Kp=300, Ki=400, Kd=80 plus nominal gravity feed-forward.
50
+ * Adaptive: RLS (λ=0.99) on [m, c, F] with state-variable filters (ω=10.0 rad/s); control
51
+ u = m̂(ωn²e − 2ζωn v + ki∫e + g) + ĉv (F̂ is estimated but not fed back because it chatters with noisy velocity), ωn=6, ζ=0.7, ki=30. The delay value is given to the controller to align its regression.
52
+ * The 0.5 m target is approached with a 1 s ramp starting at t = 0.5 s.
53
+ * Both gain sets come from grid searches of the same size and cost (IAE + 0.5·energy) on 40 tuning seeds; test seeds are disjoint.
54
+ * The browser runs a line-by-line JavaScript port; Node parity on 6 fixed runs: max abs difference 8.459899447643693e-14.
55
+
56
+ ## Usage
57
+ ```bash
58
+ python inference.py --dm 8 --delay 4 --noise 0.003 # prints metrics for both controllers
59
+ python inference.py --selftest
60
+ ```
61
+
62
+ ## Limitations
63
+ * 1-D rigid-body model; no structural flexibility, actuator dynamics beyond pure delay, or sensor bias.
64
+ * The adaptive controller is told the actuator delay; an unknown delay would need estimation.
65
+ * Results are from simulation on NVIDIA L4; no physical platform was tested.
balance_core.py ADDED
@@ -0,0 +1,120 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Next Balance: hold a vertical lift platform at its target while the payload and friction change.
2
+
3
+ Simulation only. Plant (semi-implicit Euler, dt = 0.01 s):
4
+ m(t) * (z'' + g) = u(t - delay) - c(t) * z' - F(t) * tanh(z' / 0.01)
5
+ Sensor: z + Gaussian noise. Actuator force is limited to [0, U_MAX].
6
+ Both controllers use the same alpha-beta state estimator and know the nominal payload only.
7
+
8
+ fixed PID : u = Kp e + Ki I - Kd v + m_nominal g
9
+ adaptive : recursive least squares (forgetting factor) estimates [m, c, F] from filtered
10
+ signals; u = m_hat (wn^2 e - 2 zeta wn v + ki I + g) + c_hat v
11
+ (F_hat is estimated to keep m_hat unbiased but not fed back: it chatters with noisy v)
12
+ The same code is ported line by line to web/core.js; parity is tested on fixed scenarios.
13
+ """
14
+ import math
15
+
16
+ from nlrt import Mulberry32
17
+
18
+ DT, G, U_MAX, M_PLATFORM, EPS = 0.01, 9.81, 400.0, 2.0, 0.01
19
+ ALPHA, BETA = 0.3, 0.05
20
+ Z_REF, T_STEP, T_RAMP = 0.5, 0.5, 1.0
21
+
22
+
23
+ def scenario(seed, stress=False):
24
+ r = Mulberry32(seed)
25
+ s = {"payload0": r.uniform(1, 4), "t_load": r.uniform(3.5, 4.5),
26
+ "dm": r.uniform(-1, 10 if stress else 6), "c0": r.uniform(2, 5), "F0": r.uniform(1, 3),
27
+ "t_fric": r.uniform(6.5, 7.5), "fric_gain": r.uniform(1, 6 if stress else 4),
28
+ "delay": r.randint(8 if stress else 6), "noise": r.uniform(0, 0.006 if stress else 0.003),
29
+ "T": 12.0, "seed": seed}
30
+ return s
31
+
32
+
33
+ def simulate(sc, ctrl, params, record_every=1):
34
+ rng = Mulberry32(sc["seed"] * 31 + 7)
35
+ m_nom = M_PLATFORM + sc["payload0"]
36
+ n = int(round(sc["T"] / DT))
37
+ z = v = 0.0
38
+ zh = vh = 0.0
39
+ integ = 0.0
40
+ queue = [m_nom * G] * (sc["delay"] + 1) # commands in flight; start at hover force
41
+ # adaptive state: RLS + state-variable filters
42
+ theta = [m_nom, 3.0, 2.0]
43
+ P = [[10.0, 0, 0], [0, 10.0, 0], [0, 0, 10.0]]
44
+ fz = [0.0, 0.0]
45
+ fu = [m_nom * G, 0.0]
46
+ ft = [0.0, 0.0]
47
+ w = params.get("wf", 20.0)
48
+ lam = params.get("lam", 0.995)
49
+ t_hist, z_hist, u_hist, m_hist = [], [], [], []
50
+ energy = 0.0
51
+ for k in range(n):
52
+ t = k * DT
53
+ m = M_PLATFORM + sc["payload0"] + (sc["dm"] if t >= sc["t_load"] else 0.0)
54
+ g_f = sc["fric_gain"] if t >= sc["t_fric"] else 1.0
55
+ c, F = sc["c0"] * g_f, sc["F0"] * g_f
56
+ ref = Z_REF * min(1.0, max(0.0, (t - T_STEP) / T_RAMP))
57
+ # sensor and estimator
58
+ meas = z + sc["noise"] * rng.normal()
59
+ zp = zh + vh * DT
60
+ res = meas - zp
61
+ zh = zp + ALPHA * res
62
+ vh = vh + (BETA / DT) * res
63
+ e = ref - zh
64
+ integ = max(-params.get("imax", 0.5), min(params.get("imax", 0.5), integ + e * DT))
65
+ if ctrl == "pid":
66
+ u = params["kp"] * e + params["ki"] * integ - params["kd"] * vh + m_nom * G
67
+ else:
68
+ u_applied = queue[0]
69
+ tv = math.tanh(vh / EPS)
70
+ # state-variable filters H(s) = w^2 / (s + w)^2 on measured z, applied force and friction regressor
71
+ az = w * w * (meas - fz[0]) - 2 * w * fz[1]
72
+ au = w * w * (u_applied - fu[0]) - 2 * w * fu[1]
73
+ at = w * w * (tv - ft[0]) - 2 * w * ft[1]
74
+ fz = [fz[0] + DT * fz[1], fz[1] + DT * az]
75
+ fu = [fu[0] + DT * fu[1], fu[1] + DT * au]
76
+ ft = [ft[0] + DT * ft[1], ft[1] + DT * at]
77
+ phi = [az + G, fz[1], ft[0]]
78
+ y = fu[0]
79
+ Pphi = [sum(P[i][j] * phi[j] for j in range(3)) for i in range(3)]
80
+ den = lam + sum(phi[i] * Pphi[i] for i in range(3))
81
+ K = [Pphi[i] / den for i in range(3)]
82
+ err = y - sum(phi[i] * theta[i] for i in range(3))
83
+ theta = [theta[i] + K[i] * err for i in range(3)]
84
+ P = [[(P[i][j] - K[i] * Pphi[j]) / lam for j in range(3)] for i in range(3)]
85
+ theta = [min(30.0, max(0.5, theta[0])), min(60.0, max(0.0, theta[1])), min(60.0, max(0.0, theta[2]))]
86
+ wn, zeta = params["wn"], params["zeta"]
87
+ acmd = wn * wn * e - 2 * zeta * wn * vh + params["ki_a"] * integ
88
+ u = theta[0] * (acmd + G) + theta[1] * vh
89
+ u = max(0.0, min(U_MAX, u))
90
+ queue.append(u)
91
+ u_act = queue.pop(0)
92
+ # plant
93
+ a = (u_act - c * v - F * math.tanh(v / EPS)) / m - G
94
+ v = v + a * DT
95
+ z = z + v * DT
96
+ energy += u_act * u_act * DT
97
+ if k % record_every == 0:
98
+ t_hist.append(round(t, 4))
99
+ z_hist.append(z)
100
+ u_hist.append(u_act)
101
+ m_hist.append(theta[0] if ctrl != "pid" else m_nom)
102
+ return {"t": t_hist, "z": z_hist, "u": u_hist, "m_hat": m_hist, "energy": energy}
103
+
104
+
105
+ def metrics(sc, tr, band=0.01, record_every=1):
106
+ t, z = tr["t"], tr["z"]
107
+ iae = sum(abs(z[i] - Z_REF * min(1.0, max(0.0, (t[i] - T_STEP) / T_RAMP))) for i in range(len(t))) * DT * record_every
108
+ after = [i for i in range(len(t)) if t[i] >= sc["t_load"]]
109
+ max_dev = max(abs(z[i] - Z_REF) for i in after)
110
+ last_out = None
111
+ for i in after:
112
+ if abs(z[i] - Z_REF) > band:
113
+ last_out = t[i]
114
+ settled = last_out is None or last_out < t[-1] - 0.5
115
+ settle = 0.0 if last_out is None else last_out - sc["t_load"]
116
+ step_idx = [i for i in range(len(t)) if T_STEP <= t[i] < sc["t_load"]]
117
+ overshoot = max(0.0, max(z[i] for i in step_idx) - Z_REF) if step_idx else 0.0
118
+ return {"max_deviation_m": max_dev, "settling_time_s": settle, "settled": settled,
119
+ "control_energy_kN2s": tr["energy"] / 1e6, "step_overshoot_m": overshoot,
120
+ "final_error_m": abs(z[-1] - Z_REF), "iae_ms": iae}
evaluation.json ADDED
@@ -0,0 +1,1309 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "project": "Next Balance",
3
+ "profile": "full",
4
+ "seed": 73,
5
+ "device": {
6
+ "python": "3.13.16",
7
+ "platform": "Linux-6.6.159+-x86_64-with-glibc2.39",
8
+ "cpu_count": 12,
9
+ "numpy": "2.4.6",
10
+ "torch": "2.14.1+cu130",
11
+ "cuda": true,
12
+ "gpu": "NVIDIA L4",
13
+ "colab": true
14
+ },
15
+ "data": {
16
+ "type": "simulation",
17
+ "note": "No external data. Scenarios are sampled from bc.scenario(seed) with a Mulberry32 PRNG.",
18
+ "tuning_seeds": [
19
+ 1000,
20
+ 1001,
21
+ 1002,
22
+ 1003,
23
+ 1004,
24
+ 1005,
25
+ 1006,
26
+ 1007,
27
+ 1008,
28
+ 1009,
29
+ 1010,
30
+ 1011,
31
+ 1012,
32
+ 1013,
33
+ 1014,
34
+ 1015,
35
+ 1016,
36
+ 1017,
37
+ 1018,
38
+ 1019,
39
+ 1020,
40
+ 1021,
41
+ 1022,
42
+ 1023,
43
+ 1024,
44
+ 1025,
45
+ 1026,
46
+ 1027,
47
+ 1028,
48
+ 1029,
49
+ 1030,
50
+ 1031,
51
+ 1032,
52
+ 1033,
53
+ 1034,
54
+ 1035,
55
+ 1036,
56
+ 1037,
57
+ 1038,
58
+ 1039
59
+ ],
60
+ "test_seeds": [
61
+ 5000,
62
+ 5001,
63
+ 5002,
64
+ 5003,
65
+ 5004,
66
+ 5005,
67
+ 5006,
68
+ 5007,
69
+ 5008,
70
+ 5009,
71
+ 5010,
72
+ 5011,
73
+ 5012,
74
+ 5013,
75
+ 5014,
76
+ 5015,
77
+ 5016,
78
+ 5017,
79
+ 5018,
80
+ 5019,
81
+ 5020,
82
+ 5021,
83
+ 5022,
84
+ 5023,
85
+ 5024,
86
+ 5025,
87
+ 5026,
88
+ 5027,
89
+ 5028,
90
+ 5029,
91
+ 5030,
92
+ 5031,
93
+ 5032,
94
+ 5033,
95
+ 5034,
96
+ 5035,
97
+ 5036,
98
+ 5037,
99
+ 5038,
100
+ 5039,
101
+ 5040,
102
+ 5041,
103
+ 5042,
104
+ 5043,
105
+ 5044,
106
+ 5045,
107
+ 5046,
108
+ 5047,
109
+ 5048,
110
+ 5049,
111
+ 5050,
112
+ 5051,
113
+ 5052,
114
+ 5053,
115
+ 5054,
116
+ 5055,
117
+ 5056,
118
+ 5057,
119
+ 5058,
120
+ 5059,
121
+ 5060,
122
+ 5061,
123
+ 5062,
124
+ 5063,
125
+ 5064,
126
+ 5065,
127
+ 5066,
128
+ 5067,
129
+ 5068,
130
+ 5069,
131
+ 5070,
132
+ 5071,
133
+ 5072,
134
+ 5073,
135
+ 5074,
136
+ 5075,
137
+ 5076,
138
+ 5077,
139
+ 5078,
140
+ 5079,
141
+ 5080,
142
+ 5081,
143
+ 5082,
144
+ 5083,
145
+ 5084,
146
+ 5085,
147
+ 5086,
148
+ 5087,
149
+ 5088,
150
+ 5089,
151
+ 5090,
152
+ 5091,
153
+ 5092,
154
+ 5093,
155
+ 5094,
156
+ 5095,
157
+ 5096,
158
+ 5097,
159
+ 5098,
160
+ 5099,
161
+ 5100,
162
+ 5101,
163
+ 5102,
164
+ 5103,
165
+ 5104,
166
+ 5105,
167
+ 5106,
168
+ 5107,
169
+ 5108,
170
+ 5109,
171
+ 5110,
172
+ 5111,
173
+ 5112,
174
+ 5113,
175
+ 5114,
176
+ 5115,
177
+ 5116,
178
+ 5117,
179
+ 5118,
180
+ 5119,
181
+ 5120,
182
+ 5121,
183
+ 5122,
184
+ 5123,
185
+ 5124,
186
+ 5125,
187
+ 5126,
188
+ 5127,
189
+ 5128,
190
+ 5129,
191
+ 5130,
192
+ 5131,
193
+ 5132,
194
+ 5133,
195
+ 5134,
196
+ 5135,
197
+ 5136,
198
+ 5137,
199
+ 5138,
200
+ 5139,
201
+ 5140,
202
+ 5141,
203
+ 5142,
204
+ 5143,
205
+ 5144,
206
+ 5145,
207
+ 5146,
208
+ 5147,
209
+ 5148,
210
+ 5149,
211
+ 5150,
212
+ 5151,
213
+ 5152,
214
+ 5153,
215
+ 5154,
216
+ 5155,
217
+ 5156,
218
+ 5157,
219
+ 5158,
220
+ 5159,
221
+ 5160,
222
+ 5161,
223
+ 5162,
224
+ 5163,
225
+ 5164,
226
+ 5165,
227
+ 5166,
228
+ 5167,
229
+ 5168,
230
+ 5169,
231
+ 5170,
232
+ 5171,
233
+ 5172,
234
+ 5173,
235
+ 5174,
236
+ 5175,
237
+ 5176,
238
+ 5177,
239
+ 5178,
240
+ 5179,
241
+ 5180,
242
+ 5181,
243
+ 5182,
244
+ 5183,
245
+ 5184,
246
+ 5185,
247
+ 5186,
248
+ 5187,
249
+ 5188,
250
+ 5189,
251
+ 5190,
252
+ 5191,
253
+ 5192,
254
+ 5193,
255
+ 5194,
256
+ 5195,
257
+ 5196,
258
+ 5197,
259
+ 5198,
260
+ 5199
261
+ ],
262
+ "splits": [
263
+ "nominal",
264
+ "stress"
265
+ ]
266
+ },
267
+ "controllers": {
268
+ "pid": {
269
+ "kp": 300,
270
+ "ki": 400,
271
+ "kd": 80,
272
+ "imax": 0.2
273
+ },
274
+ "adaptive": {
275
+ "wn": 6,
276
+ "zeta": 0.7,
277
+ "ki_a": 30,
278
+ "imax": 0.5,
279
+ "wf": 10.0,
280
+ "lam": 0.99
281
+ }
282
+ },
283
+ "tuning": {
284
+ "pid_grid_size": 72,
285
+ "adaptive_grid_size": 72,
286
+ "cost": "IAE + 0.5*energy on tuning seeds only"
287
+ },
288
+ "results": {
289
+ "nominal": {
290
+ "all": {
291
+ "pid": {
292
+ "n": 200,
293
+ "settled_rate": 0.81,
294
+ "max_deviation_m": {
295
+ "mean": 0.14381613039013313,
296
+ "median": 0.1275249862696633,
297
+ "p90": 0.28253078007398874
298
+ },
299
+ "settling_time_s": {
300
+ "mean": 3.3608479333645196,
301
+ "median": 1.974241643268615,
302
+ "p90": 8.032996131209657
303
+ },
304
+ "control_energy_kN2s": {
305
+ "mean": 0.06102178558463826,
306
+ "median": 0.058176729725626895,
307
+ "p90": 0.09424600123462587
308
+ },
309
+ "step_overshoot_m": {
310
+ "mean": 0.157151631590862,
311
+ "median": 0.12665297152213068,
312
+ "p90": 0.29502007387936036
313
+ },
314
+ "iae_ms": {
315
+ "mean": 0.45503818964749354,
316
+ "median": 0.34905450568001656,
317
+ "p90": 0.8012585529421845
318
+ }
319
+ },
320
+ "adaptive": {
321
+ "n": 200,
322
+ "settled_rate": 0.955,
323
+ "max_deviation_m": {
324
+ "mean": 0.14021530867902102,
325
+ "median": 0.14211990702903765,
326
+ "p90": 0.26282767141740376
327
+ },
328
+ "settling_time_s": {
329
+ "mean": 4.2852495597766715,
330
+ "median": 5.032300224225037,
331
+ "p90": 6.566111610651016
332
+ },
333
+ "control_energy_kN2s": {
334
+ "mean": 0.05169536143361678,
335
+ "median": 0.049301098669162144,
336
+ "p90": 0.08660803776277522
337
+ },
338
+ "step_overshoot_m": {
339
+ "mean": 0.06807306284281793,
340
+ "median": 0.05462651443438038,
341
+ "p90": 0.07197861751892794
342
+ },
343
+ "iae_ms": {
344
+ "mean": 0.31863218517121167,
345
+ "median": 0.3099200971622497,
346
+ "p90": 0.43473902550345433
347
+ }
348
+ }
349
+ },
350
+ "heavy_load_change(dm>4kg)": {
351
+ "pid": {
352
+ "n": 67,
353
+ "settled_rate": 0.9850746268656716,
354
+ "max_deviation_m": {
355
+ "mean": 0.1939174812147893,
356
+ "median": 0.14942781295861973,
357
+ "p90": 0.32854777427635257
358
+ },
359
+ "settling_time_s": {
360
+ "mean": 2.2086951454247887,
361
+ "median": 1.9464817980211224,
362
+ "p90": 2.57688498785533
363
+ },
364
+ "control_energy_kN2s": {
365
+ "mean": 0.08139073330549218,
366
+ "median": 0.08282572389307474,
367
+ "p90": 0.10550095041535491
368
+ },
369
+ "step_overshoot_m": {
370
+ "mean": 0.15867556885673778,
371
+ "median": 0.127260511513145,
372
+ "p90": 0.297177782280586
373
+ },
374
+ "iae_ms": {
375
+ "mean": 0.43985131775261743,
376
+ "median": 0.36080689436090807,
377
+ "p90": 0.7026169507757137
378
+ }
379
+ },
380
+ "adaptive": {
381
+ "n": 67,
382
+ "settled_rate": 1.0,
383
+ "max_deviation_m": {
384
+ "mean": 0.23871260383227924,
385
+ "median": 0.22908867132835026,
386
+ "p90": 0.3158807588752011
387
+ },
388
+ "settling_time_s": {
389
+ "mean": 5.188844399156132,
390
+ "median": 5.154934212099761,
391
+ "p90": 5.94156158191152
392
+ },
393
+ "control_energy_kN2s": {
394
+ "mean": 0.07524801163671294,
395
+ "median": 0.07448805515987396,
396
+ "p90": 0.09967525569975574
397
+ },
398
+ "step_overshoot_m": {
399
+ "mean": 0.06427099545784856,
400
+ "median": 0.05457923988808033,
401
+ "p90": 0.06674600714256958
402
+ },
403
+ "iae_ms": {
404
+ "mean": 0.3978571445007959,
405
+ "median": 0.37780565983270314,
406
+ "p90": 0.48132419208651583
407
+ }
408
+ }
409
+ },
410
+ "load_removed(dm<0)": {
411
+ "pid": {
412
+ "n": 19,
413
+ "settled_rate": 0.15789473684210525,
414
+ "max_deviation_m": {
415
+ "mean": 0.1128876127513908,
416
+ "median": 0.11739043110441866,
417
+ "p90": 0.20563845249297036
418
+ },
419
+ "settling_time_s": {
420
+ "mean": 6.875238611656486,
421
+ "median": 7.9267780026514085,
422
+ "p90": 8.43277845416218
423
+ },
424
+ "control_energy_kN2s": {
425
+ "mean": 0.03884366443411085,
426
+ "median": 0.034648360295323026,
427
+ "p90": 0.06458584784046394
428
+ },
429
+ "step_overshoot_m": {
430
+ "mean": 0.18789962168813332,
431
+ "median": 0.18525218542524624,
432
+ "p90": 0.29504986475240497
433
+ },
434
+ "iae_ms": {
435
+ "mean": 0.6300266516621532,
436
+ "median": 0.6044820004599968,
437
+ "p90": 1.0325145076139173
438
+ }
439
+ },
440
+ "adaptive": {
441
+ "n": 19,
442
+ "settled_rate": 1.0,
443
+ "max_deviation_m": {
444
+ "mean": 0.02271264795568586,
445
+ "median": 0.02211078268154465,
446
+ "p90": 0.04005277304776542
447
+ },
448
+ "settling_time_s": {
449
+ "mean": 2.042689807377756,
450
+ "median": 0.8997931448742751,
451
+ "p90": 5.115223720042035
452
+ },
453
+ "control_energy_kN2s": {
454
+ "mean": 0.021113421114524338,
455
+ "median": 0.019201658162361995,
456
+ "p90": 0.032189804794278
457
+ },
458
+ "step_overshoot_m": {
459
+ "mean": 0.06459422780946056,
460
+ "median": 0.05521344274226492,
461
+ "p90": 0.06559316956522683
462
+ },
463
+ "iae_ms": {
464
+ "mean": 0.23374347776551188,
465
+ "median": 0.2282680759822837,
466
+ "p90": 0.27913082225159824
467
+ }
468
+ }
469
+ },
470
+ "long_delay(>=40ms)": {
471
+ "pid": {
472
+ "n": 62,
473
+ "settled_rate": 0.5645161290322581,
474
+ "max_deviation_m": {
475
+ "mean": 0.2407791149488127,
476
+ "median": 0.21280761616450272,
477
+ "p90": 0.3858241371327162
478
+ },
479
+ "settling_time_s": {
480
+ "mean": 5.277618588691488,
481
+ "median": 4.957231064066291,
482
+ "p90": 8.3026737767132
483
+ },
484
+ "control_energy_kN2s": {
485
+ "mean": 0.07509517649933214,
486
+ "median": 0.07437207582494504,
487
+ "p90": 0.10165739342190952
488
+ },
489
+ "step_overshoot_m": {
490
+ "mean": 0.2675470333453268,
491
+ "median": 0.26594952405233113,
492
+ "p90": 0.36475256487452856
493
+ },
494
+ "iae_ms": {
495
+ "mean": 0.7693706105551016,
496
+ "median": 0.707538456268709,
497
+ "p90": 1.1381948605541647
498
+ }
499
+ },
500
+ "adaptive": {
501
+ "n": 62,
502
+ "settled_rate": 1.0,
503
+ "max_deviation_m": {
504
+ "mean": 0.15305024112058524,
505
+ "median": 0.14901465023109176,
506
+ "p90": 0.31620177748851447
507
+ },
508
+ "settling_time_s": {
509
+ "mean": 3.3502900485416514,
510
+ "median": 4.293903235318139,
511
+ "p90": 5.179325788370333
512
+ },
513
+ "control_energy_kN2s": {
514
+ "mean": 0.05449004408333132,
515
+ "median": 0.05690684239022806,
516
+ "p90": 0.09111935116711836
517
+ },
518
+ "step_overshoot_m": {
519
+ "mean": 0.07132464641126016,
520
+ "median": 0.05911256317436431,
521
+ "p90": 0.07594615281114882
522
+ },
523
+ "iae_ms": {
524
+ "mean": 0.32209853346149275,
525
+ "median": 0.30820227729089733,
526
+ "p90": 0.45343736201826856
527
+ }
528
+ }
529
+ },
530
+ "high_noise(>2mm)": {
531
+ "pid": {
532
+ "n": 70,
533
+ "settled_rate": 0.7857142857142857,
534
+ "max_deviation_m": {
535
+ "mean": 0.15057118202286246,
536
+ "median": 0.1293700614760638,
537
+ "p90": 0.28822695681097454
538
+ },
539
+ "settling_time_s": {
540
+ "mean": 3.643820981279282,
541
+ "median": 1.989577950113453,
542
+ "p90": 7.700501924366691
543
+ },
544
+ "control_energy_kN2s": {
545
+ "mean": 0.06362102265633562,
546
+ "median": 0.06722487878803904,
547
+ "p90": 0.0941349233520182
548
+ },
549
+ "step_overshoot_m": {
550
+ "mean": 0.159221796995918,
551
+ "median": 0.11750359601555144,
552
+ "p90": 0.29304406989731313
553
+ },
554
+ "iae_ms": {
555
+ "mean": 0.47168919775495144,
556
+ "median": 0.3544814276576983,
557
+ "p90": 0.8329923944548401
558
+ }
559
+ },
560
+ "adaptive": {
561
+ "n": 70,
562
+ "settled_rate": 0.9714285714285714,
563
+ "max_deviation_m": {
564
+ "mean": 0.14975725300086618,
565
+ "median": 0.15679079466004922,
566
+ "p90": 0.28600987282253854
567
+ },
568
+ "settling_time_s": {
569
+ "mean": 4.368072947270502,
570
+ "median": 4.987164463400841,
571
+ "p90": 5.930746293212287
572
+ },
573
+ "control_energy_kN2s": {
574
+ "mean": 0.05387462666817054,
575
+ "median": 0.05008141456365969,
576
+ "p90": 0.08583034571152345
577
+ },
578
+ "step_overshoot_m": {
579
+ "mean": 0.08508197288755587,
580
+ "median": 0.05620082236190077,
581
+ "p90": 0.2011828996843872
582
+ },
583
+ "iae_ms": {
584
+ "mean": 0.34024497787619756,
585
+ "median": 0.34166469001385996,
586
+ "p90": 0.4557991979495734
587
+ }
588
+ }
589
+ }
590
+ },
591
+ "stress": {
592
+ "all": {
593
+ "pid": {
594
+ "n": 200,
595
+ "settled_rate": 0.685,
596
+ "max_deviation_m": {
597
+ "mean": 0.2855939910188152,
598
+ "median": 0.23694032002428864,
599
+ "p90": 0.5396913399858491
600
+ },
601
+ "settling_time_s": {
602
+ "mean": 4.152803508086223,
603
+ "median": 2.1942560235317794,
604
+ "p90": 8.201717946082354
605
+ },
606
+ "control_energy_kN2s": {
607
+ "mean": 0.09809141318590711,
608
+ "median": 0.09243937319578777,
609
+ "p90": 0.16079167666575148
610
+ },
611
+ "step_overshoot_m": {
612
+ "mean": 0.2524753639024782,
613
+ "median": 0.1612935042686407,
614
+ "p90": 0.5252708190521935
615
+ },
616
+ "iae_ms": {
617
+ "mean": 0.780565929476434,
618
+ "median": 0.5146139476911349,
619
+ "p90": 1.5846599095797098
620
+ }
621
+ },
622
+ "adaptive": {
623
+ "n": 200,
624
+ "settled_rate": 0.9,
625
+ "max_deviation_m": {
626
+ "mean": 0.2923813652853192,
627
+ "median": 0.2676242979614739,
628
+ "p90": 0.5040495683093934
629
+ },
630
+ "settling_time_s": {
631
+ "mean": 5.080503508086223,
632
+ "median": 5.097949104961008,
633
+ "p90": 7.22964180780854
634
+ },
635
+ "control_energy_kN2s": {
636
+ "mean": 0.0937084449248433,
637
+ "median": 0.08360216519365156,
638
+ "p90": 0.17271182264999302
639
+ },
640
+ "step_overshoot_m": {
641
+ "mean": 0.1167212926479068,
642
+ "median": 0.06783748113317511,
643
+ "p90": 0.22595619799185618
644
+ },
645
+ "iae_ms": {
646
+ "mean": 0.5682944281041116,
647
+ "median": 0.45527211799410616,
648
+ "p90": 0.816298976476413
649
+ }
650
+ }
651
+ },
652
+ "heavy_load_change(dm>4kg)": {
653
+ "pid": {
654
+ "n": 121,
655
+ "settled_rate": 0.7272727272727273,
656
+ "max_deviation_m": {
657
+ "mean": 0.3223715079909859,
658
+ "median": 0.26211731600950616,
659
+ "p90": 0.5383691063311008
660
+ },
661
+ "settling_time_s": {
662
+ "mean": 3.8171879221668297,
663
+ "median": 2.088499198630452,
664
+ "p90": 8.09921718296595
665
+ },
666
+ "control_energy_kN2s": {
667
+ "mean": 0.12467119658494172,
668
+ "median": 0.12463517436230521,
669
+ "p90": 0.1716531663196487
670
+ },
671
+ "step_overshoot_m": {
672
+ "mean": 0.25022927158644304,
673
+ "median": 0.15479492758800617,
674
+ "p90": 0.5225685379938618
675
+ },
676
+ "iae_ms": {
677
+ "mean": 0.7151869611765325,
678
+ "median": 0.5270401524578711,
679
+ "p90": 1.436820761246725
680
+ }
681
+ },
682
+ "adaptive": {
683
+ "n": 121,
684
+ "settled_rate": 0.8760330578512396,
685
+ "max_deviation_m": {
686
+ "mean": 0.4063501081883159,
687
+ "median": 0.3568336751505972,
688
+ "p90": 0.6017598572428797
689
+ },
690
+ "settling_time_s": {
691
+ "mean": 5.438097013075922,
692
+ "median": 5.0953152170684195,
693
+ "p90": 7.720291265994311
694
+ },
695
+ "control_energy_kN2s": {
696
+ "mean": 0.12892107364240388,
697
+ "median": 0.11976973322020527,
698
+ "p90": 0.18818024615324366
699
+ },
700
+ "step_overshoot_m": {
701
+ "mean": 0.11810218105307989,
702
+ "median": 0.06371030738129291,
703
+ "p90": 0.22602373008566723
704
+ },
705
+ "iae_ms": {
706
+ "mean": 0.7057153639455523,
707
+ "median": 0.5366930024814522,
708
+ "p90": 1.2922411988463978
709
+ }
710
+ }
711
+ },
712
+ "load_removed(dm<0)": {
713
+ "pid": {
714
+ "n": 10,
715
+ "settled_rate": 0.2,
716
+ "max_deviation_m": {
717
+ "mean": 0.23407067341693608,
718
+ "median": 0.16807861473900382,
719
+ "p90": 0.575180610729667
720
+ },
721
+ "settling_time_s": {
722
+ "mean": 6.8541850189408295,
723
+ "median": 7.759394548381679,
724
+ "p90": 8.398554700994865
725
+ },
726
+ "control_energy_kN2s": {
727
+ "mean": 0.04393635494869176,
728
+ "median": 0.038042035267232294,
729
+ "p90": 0.06271953718288988
730
+ },
731
+ "step_overshoot_m": {
732
+ "mean": 0.23466321259631515,
733
+ "median": 0.22331772455169557,
734
+ "p90": 0.4657983592759367
735
+ },
736
+ "iae_ms": {
737
+ "mean": 1.0400317268749772,
738
+ "median": 0.8415718584735021,
739
+ "p90": 2.5703854149281087
740
+ }
741
+ },
742
+ "adaptive": {
743
+ "n": 10,
744
+ "settled_rate": 1.0,
745
+ "max_deviation_m": {
746
+ "mean": 0.035379577297461426,
747
+ "median": 0.025984810007543102,
748
+ "p90": 0.056359457232016325
749
+ },
750
+ "settling_time_s": {
751
+ "mean": 3.678185018940829,
752
+ "median": 3.7247554179513824,
753
+ "p90": 5.904211094054394
754
+ },
755
+ "control_energy_kN2s": {
756
+ "mean": 0.02382208855485987,
757
+ "median": 0.018928991261040865,
758
+ "p90": 0.0390841207521684
759
+ },
760
+ "step_overshoot_m": {
761
+ "mean": 0.12013487717466238,
762
+ "median": 0.08345213568486481,
763
+ "p90": 0.20912883829208387
764
+ },
765
+ "iae_ms": {
766
+ "mean": 0.2961774346218216,
767
+ "median": 0.2900404095817386,
768
+ "p90": 0.37409044056783675
769
+ }
770
+ }
771
+ },
772
+ "long_delay(>=40ms)": {
773
+ "pid": {
774
+ "n": 88,
775
+ "settled_rate": 0.5113636363636364,
776
+ "max_deviation_m": {
777
+ "mean": 0.44748223389070324,
778
+ "median": 0.3958899683793515,
779
+ "p90": 0.6711262423014366
780
+ },
781
+ "settling_time_s": {
782
+ "mean": 5.467218355154089,
783
+ "median": 6.2690533157158645,
784
+ "p90": 8.312222193563358
785
+ },
786
+ "control_energy_kN2s": {
787
+ "mean": 0.11361830217404124,
788
+ "median": 0.11070854621766016,
789
+ "p90": 0.16927245292826815
790
+ },
791
+ "step_overshoot_m": {
792
+ "mean": 0.4377477733668394,
793
+ "median": 0.31696824992869704,
794
+ "p90": 0.8382593415607009
795
+ },
796
+ "iae_ms": {
797
+ "mean": 1.2921158375655375,
798
+ "median": 0.908040805394889,
799
+ "p90": 2.6628951601532784
800
+ }
801
+ },
802
+ "adaptive": {
803
+ "n": 88,
804
+ "settled_rate": 0.8181818181818182,
805
+ "max_deviation_m": {
806
+ "mean": 0.3591045269390054,
807
+ "median": 0.3160394942419994,
808
+ "p90": 0.6877679020814136
809
+ },
810
+ "settling_time_s": {
811
+ "mean": 5.046991082426818,
812
+ "median": 4.882426600623877,
813
+ "p90": 8.007146026331931
814
+ },
815
+ "control_energy_kN2s": {
816
+ "mean": 0.10769234997335562,
817
+ "median": 0.09289250705252046,
818
+ "p90": 0.20593977630573185
819
+ },
820
+ "step_overshoot_m": {
821
+ "mean": 0.1251615146624932,
822
+ "median": 0.08383211253162448,
823
+ "p90": 0.2130355115166301
824
+ },
825
+ "iae_ms": {
826
+ "mean": 0.738068806680679,
827
+ "median": 0.5295870522696209,
828
+ "p90": 1.610208459273406
829
+ }
830
+ }
831
+ },
832
+ "high_noise(>2mm)": {
833
+ "pid": {
834
+ "n": 143,
835
+ "settled_rate": 0.6503496503496503,
836
+ "max_deviation_m": {
837
+ "mean": 0.29415987011218553,
838
+ "median": 0.2380274981292655,
839
+ "p90": 0.5495332120095521
840
+ },
841
+ "settling_time_s": {
842
+ "mean": 4.362653108898655,
843
+ "median": 2.301162269562483,
844
+ "p90": 8.214102686112748
845
+ },
846
+ "control_energy_kN2s": {
847
+ "mean": 0.10091302880053449,
848
+ "median": 0.09599427160852374,
849
+ "p90": 0.1607164639012419
850
+ },
851
+ "step_overshoot_m": {
852
+ "mean": 0.26364479310089656,
853
+ "median": 0.16788356754574418,
854
+ "p90": 0.6146570039239294
855
+ },
856
+ "iae_ms": {
857
+ "mean": 0.8180334981865162,
858
+ "median": 0.5380534841109816,
859
+ "p90": 1.6515277272136497
860
+ }
861
+ },
862
+ "adaptive": {
863
+ "n": 143,
864
+ "settled_rate": 0.8951048951048951,
865
+ "max_deviation_m": {
866
+ "mean": 0.30965762148192066,
867
+ "median": 0.2777016596181938,
868
+ "p90": 0.5262971253416756
869
+ },
870
+ "settling_time_s": {
871
+ "mean": 5.187688073933619,
872
+ "median": 5.100582992853596,
873
+ "p90": 7.224360503148287
874
+ },
875
+ "control_energy_kN2s": {
876
+ "mean": 0.09728914269230954,
877
+ "median": 0.08675591607064322,
878
+ "p90": 0.17717254306391475
879
+ },
880
+ "step_overshoot_m": {
881
+ "mean": 0.13831806715781195,
882
+ "median": 0.10466612578841261,
883
+ "p90": 0.23406707310752048
884
+ },
885
+ "iae_ms": {
886
+ "mean": 0.615803499784918,
887
+ "median": 0.5035751254105679,
888
+ "p90": 1.0919949579896426
889
+ }
890
+ }
891
+ }
892
+ }
893
+ },
894
+ "paired_wins": {
895
+ "nominal": {
896
+ "max_deviation_m": {
897
+ "adaptive_better": 86,
898
+ "n": 200
899
+ },
900
+ "settling_time_s": {
901
+ "adaptive_better": 62,
902
+ "n": 200
903
+ },
904
+ "control_energy_kN2s": {
905
+ "adaptive_better": 178,
906
+ "n": 200
907
+ }
908
+ },
909
+ "stress": {
910
+ "max_deviation_m": {
911
+ "adaptive_better": 69,
912
+ "n": 200
913
+ },
914
+ "settling_time_s": {
915
+ "adaptive_better": 67,
916
+ "n": 200
917
+ },
918
+ "control_energy_kN2s": {
919
+ "adaptive_better": 148,
920
+ "n": 200
921
+ }
922
+ }
923
+ },
924
+ "failures_not_settled": [
925
+ {
926
+ "split": "stress",
927
+ "seed": 5006,
928
+ "ctrl": "pid",
929
+ "dm": 9.980523464269936,
930
+ "delay": 7,
931
+ "noise": 0.0001678999112918973,
932
+ "fric_gain": 3.147387076402083,
933
+ "max_deviation_m": 1.4955338006835528,
934
+ "settling_time_s": 7.829327588910237,
935
+ "settled": false,
936
+ "control_energy_kN2s": 0.19012256847329378,
937
+ "step_overshoot_m": 1.531371510018829,
938
+ "final_error_m": 0.06089847967813927,
939
+ "iae_ms": 3.6463211724444267
940
+ },
941
+ {
942
+ "split": "stress",
943
+ "seed": 5023,
944
+ "ctrl": "pid",
945
+ "dm": 0.6660738771315664,
946
+ "delay": 7,
947
+ "noise": 0.0033733501229435206,
948
+ "fric_gain": 1.198689583921805,
949
+ "max_deviation_m": 1.4831441445059361,
950
+ "settling_time_s": 8.359261705316603,
951
+ "settled": false,
952
+ "control_energy_kN2s": 0.07519035099964805,
953
+ "step_overshoot_m": 1.6284352896230545,
954
+ "final_error_m": 0.05351668852134883,
955
+ "iae_ms": 8.73299760011718
956
+ },
957
+ {
958
+ "split": "stress",
959
+ "seed": 5096,
960
+ "ctrl": "adaptive",
961
+ "dm": 7.563551286933944,
962
+ "delay": 7,
963
+ "noise": 0.003959082144778222,
964
+ "fric_gain": 2.2102824868634343,
965
+ "max_deviation_m": 1.316670825257937,
966
+ "settling_time_s": 8.486538743143901,
967
+ "settled": false,
968
+ "control_energy_kN2s": 0.20462467580768454,
969
+ "step_overshoot_m": 0.14727423074421797,
970
+ "final_error_m": 0.13277760877616462,
971
+ "iae_ms": 2.488541381759503
972
+ },
973
+ {
974
+ "split": "stress",
975
+ "seed": 5175,
976
+ "ctrl": "adaptive",
977
+ "dm": 6.1426523886621,
978
+ "delay": 7,
979
+ "noise": 0.0053911508638411765,
980
+ "fric_gain": 1.9120526164770126,
981
+ "max_deviation_m": 1.2476126604599047,
982
+ "settling_time_s": 7.529022064302117,
983
+ "settled": false,
984
+ "control_energy_kN2s": 0.17456117867565046,
985
+ "step_overshoot_m": 0.6249351654551218,
986
+ "final_error_m": 0.03972174106661125,
987
+ "iae_ms": 3.5915506243342383
988
+ },
989
+ {
990
+ "split": "stress",
991
+ "seed": 5006,
992
+ "ctrl": "adaptive",
993
+ "dm": 9.980523464269936,
994
+ "delay": 7,
995
+ "noise": 0.0001678999112918973,
996
+ "fric_gain": 3.147387076402083,
997
+ "max_deviation_m": 1.1536597679275231,
998
+ "settling_time_s": 7.829327588910237,
999
+ "settled": false,
1000
+ "control_energy_kN2s": 0.26824041120762354,
1001
+ "step_overshoot_m": 0.11207839091492533,
1002
+ "final_error_m": 0.019867886094438014,
1003
+ "iae_ms": 2.1177172781242066
1004
+ },
1005
+ {
1006
+ "split": "stress",
1007
+ "seed": 5195,
1008
+ "ctrl": "adaptive",
1009
+ "dm": 9.22272419044748,
1010
+ "delay": 7,
1011
+ "noise": 0.0058828911166638135,
1012
+ "fric_gain": 1.194676574319601,
1013
+ "max_deviation_m": 1.0659340221835816,
1014
+ "settling_time_s": 7.894380318708718,
1015
+ "settled": false,
1016
+ "control_energy_kN2s": 0.3454769769300227,
1017
+ "step_overshoot_m": 0.46184282776162233,
1018
+ "final_error_m": 0.1655872270968045,
1019
+ "iae_ms": 2.7988479737272667
1020
+ },
1021
+ {
1022
+ "split": "stress",
1023
+ "seed": 5133,
1024
+ "ctrl": "pid",
1025
+ "dm": 1.8998408154584467,
1026
+ "delay": 7,
1027
+ "noise": 0.0026589490757323803,
1028
+ "fric_gain": 4.598371108528227,
1029
+ "max_deviation_m": 1.0343645974383842,
1030
+ "settling_time_s": 8.154892178261653,
1031
+ "settled": false,
1032
+ "control_energy_kN2s": 0.08827018005550984,
1033
+ "step_overshoot_m": 1.033778911690269,
1034
+ "final_error_m": 0.03261518791343887,
1035
+ "iae_ms": 2.8476171062120263
1036
+ },
1037
+ {
1038
+ "split": "stress",
1039
+ "seed": 5100,
1040
+ "ctrl": "pid",
1041
+ "dm": 3.6696945358999074,
1042
+ "delay": 7,
1043
+ "noise": 0.0011527455830946565,
1044
+ "fric_gain": 4.752193516120315,
1045
+ "max_deviation_m": 0.9756993377577181,
1046
+ "settling_time_s": 7.547322923792526,
1047
+ "settled": false,
1048
+ "control_energy_kN2s": 0.09588222903427067,
1049
+ "step_overshoot_m": 0.9049223352851041,
1050
+ "final_error_m": 0.0685738247125311,
1051
+ "iae_ms": 2.894819580123207
1052
+ },
1053
+ {
1054
+ "split": "stress",
1055
+ "seed": 5102,
1056
+ "ctrl": "pid",
1057
+ "dm": 0.6093599193263799,
1058
+ "delay": 7,
1059
+ "noise": 0.005132095199078321,
1060
+ "fric_gain": 2.254251067293808,
1061
+ "max_deviation_m": 0.9170366202760085,
1062
+ "settling_time_s": 7.612362380614504,
1063
+ "settled": false,
1064
+ "control_energy_kN2s": 0.07657289572972086,
1065
+ "step_overshoot_m": 0.8800238157090448,
1066
+ "final_error_m": 0.19899685234966702,
1067
+ "iae_ms": 3.367243492551143
1068
+ },
1069
+ {
1070
+ "split": "stress",
1071
+ "seed": 5085,
1072
+ "ctrl": "adaptive",
1073
+ "dm": 9.970788925420493,
1074
+ "delay": 6,
1075
+ "noise": 0.003966700122691691,
1076
+ "fric_gain": 3.480353902792558,
1077
+ "max_deviation_m": 0.8416004299145812,
1078
+ "settling_time_s": 7.963486954141409,
1079
+ "settled": false,
1080
+ "control_energy_kN2s": 0.2674837837657171,
1081
+ "step_overshoot_m": 0.0928410096215192,
1082
+ "final_error_m": 0.007123170747089735,
1083
+ "iae_ms": 1.3334591852754718
1084
+ },
1085
+ {
1086
+ "split": "stress",
1087
+ "seed": 5157,
1088
+ "ctrl": "adaptive",
1089
+ "dm": 8.91988121997565,
1090
+ "delay": 7,
1091
+ "noise": 0.0031509445421397685,
1092
+ "fric_gain": 1.5880136415362358,
1093
+ "max_deviation_m": 0.7961974430793823,
1094
+ "settling_time_s": 8.317174410689622,
1095
+ "settled": false,
1096
+ "control_energy_kN2s": 0.2309670192260483,
1097
+ "step_overshoot_m": 0.10466612578841261,
1098
+ "final_error_m": 0.02264756800309564,
1099
+ "iae_ms": 1.7598109808020048
1100
+ },
1101
+ {
1102
+ "split": "stress",
1103
+ "seed": 5059,
1104
+ "ctrl": "pid",
1105
+ "dm": 4.724308278411627,
1106
+ "delay": 7,
1107
+ "noise": 0.0024085854361765085,
1108
+ "fric_gain": 3.715345413889736,
1109
+ "max_deviation_m": 0.7907129530513459,
1110
+ "settling_time_s": 7.909283031253144,
1111
+ "settled": false,
1112
+ "control_energy_kN2s": 0.12761404319137074,
1113
+ "step_overshoot_m": 1.080833673349849,
1114
+ "final_error_m": 0.10022215626914965,
1115
+ "iae_ms": 2.6648132729636957
1116
+ },
1117
+ {
1118
+ "split": "stress",
1119
+ "seed": 5016,
1120
+ "ctrl": "pid",
1121
+ "dm": -0.7718999427743256,
1122
+ "delay": 5,
1123
+ "noise": 0.005794490132946521,
1124
+ "fric_gain": 2.2526944710407406,
1125
+ "max_deviation_m": 0.7874831213894129,
1126
+ "settling_time_s": 7.621718591796235,
1127
+ "settled": false,
1128
+ "control_energy_kN2s": 0.0284492735268269,
1129
+ "step_overshoot_m": 0.4707945858910141,
1130
+ "final_error_m": 0.0037102530621222485,
1131
+ "iae_ms": 2.560197893052792
1132
+ },
1133
+ {
1134
+ "split": "stress",
1135
+ "seed": 5142,
1136
+ "ctrl": "adaptive",
1137
+ "dm": 8.864927280694246,
1138
+ "delay": 7,
1139
+ "noise": 0.004657371095381677,
1140
+ "fric_gain": 2.6148215322755277,
1141
+ "max_deviation_m": 0.744946802372582,
1142
+ "settling_time_s": 8.036368309762329,
1143
+ "settled": false,
1144
+ "control_energy_kN2s": 0.24463553623945897,
1145
+ "step_overshoot_m": 0.11217331513372353,
1146
+ "final_error_m": 0.1136988270425372,
1147
+ "iae_ms": 1.6716054537789706
1148
+ },
1149
+ {
1150
+ "split": "stress",
1151
+ "seed": 5195,
1152
+ "ctrl": "pid",
1153
+ "dm": 9.22272419044748,
1154
+ "delay": 7,
1155
+ "noise": 0.0058828911166638135,
1156
+ "fric_gain": 1.194676574319601,
1157
+ "max_deviation_m": 0.6989776707091494,
1158
+ "settling_time_s": 7.894380318708718,
1159
+ "settled": false,
1160
+ "control_energy_kN2s": 0.20964382807403195,
1161
+ "step_overshoot_m": 0.500946409112929,
1162
+ "final_error_m": 0.03501792355889133,
1163
+ "iae_ms": 1.4919790155406412
1164
+ },
1165
+ {
1166
+ "split": "stress",
1167
+ "seed": 5148,
1168
+ "ctrl": "pid",
1169
+ "dm": 0.36287928046658635,
1170
+ "delay": 7,
1171
+ "noise": 0.0021170607511885463,
1172
+ "fric_gain": 3.3758450464811176,
1173
+ "max_deviation_m": 0.6591899158409884,
1174
+ "settling_time_s": 7.522007108675316,
1175
+ "settled": false,
1176
+ "control_energy_kN2s": 0.10025643302486421,
1177
+ "step_overshoot_m": 0.7136313896317514,
1178
+ "final_error_m": 0.19634350996526406,
1179
+ "iae_ms": 2.687511950654677
1180
+ },
1181
+ {
1182
+ "split": "stress",
1183
+ "seed": 5142,
1184
+ "ctrl": "pid",
1185
+ "dm": 8.864927280694246,
1186
+ "delay": 7,
1187
+ "noise": 0.004657371095381677,
1188
+ "fric_gain": 2.6148215322755277,
1189
+ "max_deviation_m": 0.6439192597179966,
1190
+ "settling_time_s": 8.036368309762329,
1191
+ "settled": false,
1192
+ "control_energy_kN2s": 0.1711354579845949,
1193
+ "step_overshoot_m": 0.7740934650765139,
1194
+ "final_error_m": 0.022579959199304833,
1195
+ "iae_ms": 1.8425790706872633
1196
+ },
1197
+ {
1198
+ "split": "stress",
1199
+ "seed": 5061,
1200
+ "ctrl": "pid",
1201
+ "dm": 9.228373318677768,
1202
+ "delay": 6,
1203
+ "noise": 0.0011696923300623895,
1204
+ "fric_gain": 5.410471260780469,
1205
+ "max_deviation_m": 0.6325691107838162,
1206
+ "settling_time_s": 8.460148148247972,
1207
+ "settled": false,
1208
+ "control_energy_kN2s": 0.18089816930252955,
1209
+ "step_overshoot_m": 0.38499176730395857,
1210
+ "final_error_m": 0.035145666346633964,
1211
+ "iae_ms": 1.2513241669549189
1212
+ },
1213
+ {
1214
+ "split": "stress",
1215
+ "seed": 5157,
1216
+ "ctrl": "pid",
1217
+ "dm": 8.91988121997565,
1218
+ "delay": 7,
1219
+ "noise": 0.0031509445421397685,
1220
+ "fric_gain": 1.5880136415362358,
1221
+ "max_deviation_m": 0.6260006043022233,
1222
+ "settling_time_s": 8.317174410689622,
1223
+ "settled": false,
1224
+ "control_energy_kN2s": 0.15394873113543087,
1225
+ "step_overshoot_m": 0.7700712765468121,
1226
+ "final_error_m": 0.024501909634130847,
1227
+ "iae_ms": 1.6649193259750243
1228
+ },
1229
+ {
1230
+ "split": "stress",
1231
+ "seed": 5179,
1232
+ "ctrl": "pid",
1233
+ "dm": 1.4444636353291571,
1234
+ "delay": 7,
1235
+ "noise": 0.005041373414453119,
1236
+ "fric_gain": 3.1350709605030715,
1237
+ "max_deviation_m": 0.6189971076759178,
1238
+ "settling_time_s": 7.834829899957404,
1239
+ "settled": false,
1240
+ "control_energy_kN2s": 0.06073641417786291,
1241
+ "step_overshoot_m": 0.9190520239276558,
1242
+ "final_error_m": 0.057428660944872756,
1243
+ "iae_ms": 3.1538552150389405
1244
+ }
1245
+ ],
1246
+ "n_not_settled": {
1247
+ "nominal/pid": 38,
1248
+ "nominal/adaptive": 9,
1249
+ "stress/pid": 63,
1250
+ "stress/adaptive": 20
1251
+ },
1252
+ "media": {
1253
+ "file": "replay.mp4",
1254
+ "frames": 600,
1255
+ "fps": 25,
1256
+ "bytes": 182074
1257
+ },
1258
+ "tests": {
1259
+ "reload": {
1260
+ "returncode": 0,
1261
+ "seconds": 0.26,
1262
+ "process": "fresh python -E subprocess, cwd = temporary copy of the model folder",
1263
+ "passed": true,
1264
+ "cases": 6,
1265
+ "max_abs_diff": 0.0
1266
+ },
1267
+ "node_parity": {
1268
+ "cases": 6,
1269
+ "failed": 0,
1270
+ "maxDiff": 8.459899447643693e-14,
1271
+ "results": [
1272
+ {
1273
+ "name": "case0-pid",
1274
+ "maxDiff": 1.1102230246251565e-16
1275
+ },
1276
+ {
1277
+ "name": "case0-adaptive",
1278
+ "maxDiff": 7.716050021144838e-15
1279
+ },
1280
+ {
1281
+ "name": "case1-pid",
1282
+ "maxDiff": 1.3322676295501878e-15
1283
+ },
1284
+ {
1285
+ "name": "case1-adaptive",
1286
+ "maxDiff": 8.459899447643693e-14
1287
+ },
1288
+ {
1289
+ "name": "case2-pid",
1290
+ "maxDiff": 1.1102230246251565e-16
1291
+ },
1292
+ {
1293
+ "name": "case2-adaptive",
1294
+ "maxDiff": 7.494005416219807e-16
1295
+ }
1296
+ ],
1297
+ "ms": 29,
1298
+ "runtime": "node v22.23.4",
1299
+ "status": "passed"
1300
+ },
1301
+ "browser": {
1302
+ "status": "skipped",
1303
+ "reason": "playwright not installed"
1304
+ }
1305
+ },
1306
+ "not_run": [
1307
+ "No hardware experiment. Actuator, friction and sensor models are idealised."
1308
+ ]
1309
+ }
inference.py ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Next Balance: run the tuned fixed-PID and adaptive controllers on a scenario.
2
+
3
+ Usage:
4
+ python inference.py --selftest
5
+ python inference.py --dm 8 --delay 4 --noise 0.003
6
+ """
7
+ import argparse
8
+ import json
9
+ import os
10
+ import sys
11
+
12
+ HERE = os.path.dirname(os.path.abspath(__file__))
13
+ sys.path.insert(0, HERE)
14
+
15
+ import balance_core as bc # noqa: E402
16
+
17
+
18
+ def load(path=os.path.join(HERE, "model.json")):
19
+ with open(path, encoding="utf-8") as f:
20
+ return json.load(f)["meta"]["params"]
21
+
22
+
23
+ def run(scenario, controller="adaptive", params=None):
24
+ params = params or load()
25
+ tr = bc.simulate(scenario, controller, params[controller])
26
+ return bc.metrics(scenario, tr), tr
27
+
28
+
29
+ def _selftest():
30
+ params = load()
31
+ with open(os.path.join(HERE, "selftest.json"), encoding="utf-8") as f:
32
+ cases = json.load(f)["cases"]
33
+ worst = 0.0
34
+ for c in cases:
35
+ m, _ = run(c["scenario"], c["ctrl"], params)
36
+ for k, v in c["expected"].items():
37
+ if isinstance(v, bool):
38
+ assert m[k] == v, (k, m[k], v)
39
+ else:
40
+ worst = max(worst, abs(m[k] - v))
41
+ ok = worst < 1e-9
42
+ print(json.dumps({"passed": ok, "cases": len(cases), "max_abs_diff": worst}))
43
+ return ok
44
+
45
+
46
+ if __name__ == "__main__":
47
+ ap = argparse.ArgumentParser()
48
+ ap.add_argument("--selftest", action="store_true")
49
+ ap.add_argument("--payload0", type=float, default=2.5)
50
+ ap.add_argument("--dm", type=float, default=5.0)
51
+ ap.add_argument("--fric_gain", type=float, default=2.5)
52
+ ap.add_argument("--delay", type=int, default=3, help="actuator delay in 10 ms steps")
53
+ ap.add_argument("--noise", type=float, default=0.002, help="sensor noise std in m")
54
+ ap.add_argument("--seed", type=int, default=7)
55
+ a = ap.parse_args()
56
+ if a.selftest:
57
+ sys.exit(0 if _selftest() else 1)
58
+ sc = {"payload0": a.payload0, "t_load": 4.0, "dm": a.dm, "c0": 3.5, "F0": 2.0, "t_fric": 7.0,
59
+ "fric_gain": a.fric_gain, "delay": a.delay, "noise": a.noise, "T": 12.0, "seed": a.seed}
60
+ for ctrl in ("pid", "adaptive"):
61
+ print(ctrl, json.dumps(run(sc, ctrl)[0]))
model.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"format":"nlrt-1","meta":{"name":"Next Balance","params":{"pid":{"kp":300,"ki":400,"kd":80,"imax":0.2},"adaptive":{"wn":6,"zeta":0.7,"ki_a":30,"imax":0.5,"wf":10.0,"lam":0.99}},"dt":0.01,"u_max":400.0,"platform_mass":2.0,"target_m":0.5,"selection":"grid search on tuning seeds 1000.., cost IAE + 0.5*energy"},"nets":{},"tensors":{}}
nlrt.py ADDED
@@ -0,0 +1,226 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """nlrt: tiny portable runtime shared by Next Lab models.
2
+
3
+ The same layer list runs in NumPy (this file) and in the browser (nlrt.js).
4
+ Weights are stored as base64 little-endian float32 inside model.json.
5
+ Only NumPy is required to load and run a model.
6
+ """
7
+ import base64
8
+ import json
9
+ import math
10
+
11
+ import numpy as np
12
+
13
+ __all__ = ["b64e", "b64d", "forward", "load_model", "Mulberry32", "export_sequential"]
14
+
15
+
16
+ def b64e(a):
17
+ return base64.b64encode(np.ascontiguousarray(a, dtype="<f4").tobytes()).decode("ascii")
18
+
19
+
20
+ def b64d(s, shape=None):
21
+ a = np.frombuffer(base64.b64decode(s), dtype="<f4").astype(np.float64)
22
+ return a.reshape(shape) if shape is not None else a
23
+
24
+
25
+ def load_model(path):
26
+ with open(path, "r", encoding="utf-8") as f:
27
+ spec = json.load(f)
28
+ for net in spec.get("nets", {}).values():
29
+ _decode(net)
30
+ for k, v in list(spec.get("tensors", {}).items()):
31
+ spec["tensors"][k] = b64d(v["data"], v["shape"])
32
+ return spec
33
+
34
+
35
+ def _decode(layers):
36
+ for L in layers:
37
+ for key in ("w", "b"):
38
+ if key in L and isinstance(L[key], str):
39
+ L[key] = b64d(L[key], L.get(key + "_shape"))
40
+
41
+
42
+ class Mulberry32:
43
+ """Deterministic PRNG with an identical JavaScript twin in nlrt.js."""
44
+
45
+ def __init__(self, seed):
46
+ self.a = seed & 0xFFFFFFFF
47
+
48
+ def random(self):
49
+ self.a = (self.a + 0x6D2B79F5) & 0xFFFFFFFF
50
+ t = self.a
51
+ t = ((t ^ (t >> 15)) * (t | 1)) & 0xFFFFFFFF
52
+ t ^= (t + (((t ^ (t >> 7)) * (t | 61)) & 0xFFFFFFFF)) & 0xFFFFFFFF
53
+ return ((t ^ (t >> 14)) & 0xFFFFFFFF) / 4294967296.0
54
+
55
+ def uniform(self, lo, hi):
56
+ return lo + (hi - lo) * self.random()
57
+
58
+ def randint(self, n):
59
+ return int(math.floor(self.random() * n))
60
+
61
+ def normal(self):
62
+ u = max(self.random(), 1e-12)
63
+ v = self.random()
64
+ return math.sqrt(-2.0 * math.log(u)) * math.cos(2.0 * math.pi * v)
65
+
66
+
67
+ def _pair(v):
68
+ return list(v) if isinstance(v, (list, tuple)) else [v, v]
69
+
70
+
71
+ def _conv2d(x, w, b, stride, pad):
72
+ n, c, h, wd = x.shape
73
+ o, _, kh, kw = w.shape
74
+ sh, sw = stride
75
+ ph, pw = pad
76
+ xp = np.pad(x, ((0, 0), (0, 0), (ph, ph), (pw, pw)))
77
+ oh = (h + 2 * ph - kh) // sh + 1
78
+ ow = (wd + 2 * pw - kw) // sw + 1
79
+ cols = np.empty((n, c, kh, kw, oh, ow), dtype=x.dtype)
80
+ for i in range(kh):
81
+ for j in range(kw):
82
+ cols[:, :, i, j] = xp[:, :, i:i + sh * oh:sh, j:j + sw * ow:sw]
83
+ out = np.tensordot(cols, w, axes=([1, 2, 3], [1, 2, 3])) # n, oh, ow, o
84
+ out = out.transpose(0, 3, 1, 2)
85
+ if b is not None:
86
+ out = out + b[None, :, None, None]
87
+ return out
88
+
89
+
90
+ def _maxpool(x, k):
91
+ kh, kw = k
92
+ n, c, h, w = x.shape
93
+ oh, ow = h // kh, w // kw
94
+ x = x[:, :, :oh * kh, :ow * kw].reshape(n, c, oh, kh, ow, kw)
95
+ return x.max(axis=(3, 5))
96
+
97
+
98
+ def forward(layers, x):
99
+ """Run a layer list on a batch. x: (N, D) or (N, C, H, W) or (N, C, L)."""
100
+ x = np.asarray(x, dtype=np.float64)
101
+ for L in layers:
102
+ t = L["type"]
103
+ if t == "linear":
104
+ x = x @ L["w"].T + L["b"]
105
+ elif t == "conv2d":
106
+ squeeze = x.ndim == 3
107
+ if squeeze:
108
+ x = x[:, :, None, :]
109
+ x = _conv2d(x, L["w"], L["b"], _pair(L["stride"]), _pair(L["pad"]))
110
+ if squeeze:
111
+ x = x[:, :, 0, :]
112
+ elif t == "relu":
113
+ x = np.maximum(x, 0)
114
+ elif t == "tanh":
115
+ x = np.tanh(x)
116
+ elif t == "sigmoid":
117
+ x = 1 / (1 + np.exp(-x))
118
+ elif t == "maxpool":
119
+ squeeze = x.ndim == 3
120
+ if squeeze:
121
+ x = x[:, :, None, :]
122
+ x = _maxpool(x, _pair(L["k"]))
123
+ if squeeze:
124
+ x = x[:, :, 0, :]
125
+ elif t == "gap":
126
+ x = x.reshape(x.shape[0], x.shape[1], -1).mean(axis=2)
127
+ elif t == "flatten":
128
+ x = x.reshape(x.shape[0], -1)
129
+ elif t == "upsample":
130
+ x = x.repeat(L["s"], axis=-1)
131
+ if x.ndim == 4:
132
+ x = x.repeat(L["s"], axis=-2)
133
+ elif t == "l2norm":
134
+ x = x / np.maximum(np.linalg.norm(x, axis=1, keepdims=True), 1e-12)
135
+ elif t == "reshape":
136
+ x = x.reshape([x.shape[0]] + list(L["shape"]))
137
+ else:
138
+ raise ValueError("unknown layer " + t)
139
+ return x
140
+
141
+
142
+ def export_sequential(seq):
143
+ """Convert a torch nn.Sequential into a portable layer list (BatchNorm folded)."""
144
+ import torch.nn as nn
145
+
146
+ mods = list(seq.modules())[1:] if not isinstance(seq, (list, tuple)) else list(seq)
147
+ mods = [m for m in mods if not isinstance(m, nn.Sequential)]
148
+ layers = []
149
+
150
+ def tensor(t):
151
+ return t.detach().cpu().double().numpy()
152
+
153
+ for m in mods:
154
+ if isinstance(m, nn.Linear):
155
+ b = tensor(m.bias) if m.bias is not None else np.zeros(m.out_features)
156
+ layers.append({"type": "linear", "w": tensor(m.weight), "b": b})
157
+ elif isinstance(m, (nn.Conv2d, nn.Conv1d)):
158
+ w = tensor(m.weight)
159
+ b = tensor(m.bias) if m.bias is not None else np.zeros(w.shape[0])
160
+ if isinstance(m, nn.Conv1d):
161
+ w = w[:, :, None, :]
162
+ stride, pad = [1, m.stride[0]], [0, m.padding[0]]
163
+ else:
164
+ stride, pad = list(m.stride), list(m.padding)
165
+ layers.append({"type": "conv2d", "w": w, "b": b, "stride": stride, "pad": pad})
166
+ elif isinstance(m, (nn.BatchNorm1d, nn.BatchNorm2d)):
167
+ prev = layers[-1]
168
+ assert prev["type"] in ("conv2d", "linear"), "BatchNorm must follow conv/linear"
169
+ scale = tensor(m.weight) / np.sqrt(tensor(m.running_var) + m.eps)
170
+ shift = tensor(m.bias) - tensor(m.running_mean) * scale
171
+ if prev["type"] == "conv2d":
172
+ prev["w"] = prev["w"] * scale[:, None, None, None]
173
+ else:
174
+ prev["w"] = prev["w"] * scale[:, None]
175
+ prev["b"] = prev["b"] * scale + shift
176
+ elif isinstance(m, nn.ReLU):
177
+ layers.append({"type": "relu"})
178
+ elif isinstance(m, nn.Tanh):
179
+ layers.append({"type": "tanh"})
180
+ elif isinstance(m, nn.Sigmoid):
181
+ layers.append({"type": "sigmoid"})
182
+ elif isinstance(m, nn.MaxPool2d):
183
+ layers.append({"type": "maxpool", "k": _pair(m.kernel_size)})
184
+ elif isinstance(m, nn.MaxPool1d):
185
+ k = m.kernel_size if isinstance(m, int) else m.kernel_size
186
+ layers.append({"type": "maxpool", "k": [1, k if isinstance(k, int) else k[0]]})
187
+ elif isinstance(m, (nn.AdaptiveAvgPool2d, nn.AdaptiveAvgPool1d)):
188
+ layers.append({"type": "gap"})
189
+ elif isinstance(m, nn.Flatten):
190
+ layers.append({"type": "flatten"})
191
+ elif isinstance(m, nn.Upsample):
192
+ layers.append({"type": "upsample", "s": int(m.scale_factor)})
193
+ elif isinstance(m, nn.Dropout):
194
+ continue
195
+ elif m.__class__.__name__ == "L2Norm":
196
+ layers.append({"type": "l2norm"})
197
+ elif m.__class__.__name__ == "Reshape":
198
+ layers.append({"type": "reshape", "shape": list(m.shape)})
199
+ else:
200
+ raise ValueError("cannot export " + m.__class__.__name__)
201
+ return layers
202
+
203
+
204
+ def encode_layers(layers):
205
+ out = []
206
+ for L in layers:
207
+ L2 = dict(L)
208
+ for key in ("w", "b"):
209
+ if key in L2:
210
+ arr = np.asarray(L2[key])
211
+ L2[key + "_shape"] = list(arr.shape)
212
+ L2[key] = b64e(arr)
213
+ out.append(L2)
214
+ return out
215
+
216
+
217
+ def save_model(path, nets=None, tensors=None, meta=None):
218
+ spec = {"format": "nlrt-1", "meta": meta or {}, "nets": {}, "tensors": {}}
219
+ for name, layers in (nets or {}).items():
220
+ spec["nets"][name] = encode_layers(layers)
221
+ for name, arr in (tensors or {}).items():
222
+ arr = np.asarray(arr)
223
+ spec["tensors"][name] = {"shape": list(arr.shape), "data": b64e(arr)}
224
+ with open(path, "w", encoding="utf-8") as f:
225
+ json.dump(spec, f, separators=(",", ":"))
226
+ return spec
preview.png ADDED

Git LFS Details

  • SHA256: c5128428491d21f444421549f7f632a9418abf899f301a17e1bf1aceb7b04c14
  • Pointer size: 131 Bytes
  • Size of remote file: 106 kB
replay.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca7bc3f4c862bd01802435a15b075ef0f2f46cca1db7794e03493b4cf4f0a81f
3
+ size 182074
selftest.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"cases": [{"scenario": {"payload0": 2.5, "t_load": 4.0, "dm": 5.0, "c0": 3.5, "F0": 2.0, "t_fric": 7.0, "fric_gain": 2.5, "delay": 3, "noise": 0.002, "T": 12.0, "seed": 7}, "ctrl": "pid", "expected": {"max_deviation_m": 0.21246549316426738, "settling_time_s": 1.9100000000000001, "settled": true, "control_energy_kN2s": 0.08585167617097979, "step_overshoot_m": 0.16121816396249167, "final_error_m": 0.0012782173124000207, "iae_ms": 0.42688203186966583}}, {"scenario": {"payload0": 2.5, "t_load": 4.0, "dm": 5.0, "c0": 3.5, "F0": 2.0, "t_fric": 7.0, "fric_gain": 2.5, "delay": 3, "noise": 0.002, "T": 12.0, "seed": 7}, "ctrl": "adaptive", "expected": {"max_deviation_m": 0.2399659109836748, "settling_time_s": 5.130000000000001, "settled": true, "control_energy_kN2s": 0.07724842430435602, "step_overshoot_m": 0.054537633646371764, "final_error_m": 0.002987176552042059, "iae_ms": 0.37910133631047316}}, {"scenario": {"payload0": 1.2, "t_load": 4.0, "dm": 11.0, "c0": 3.5, "F0": 2.0, "t_fric": 7.0, "fric_gain": 5.0, "delay": 6, "noise": 0.006, "T": 12.0, "seed": 3}, "ctrl": "pid", "expected": {"max_deviation_m": 0.6834092876786254, "settling_time_s": 7.99, "settled": false, "control_energy_kN2s": 0.19103139746244666, "step_overshoot_m": 0.7949034817694149, "final_error_m": 0.09350074369439826, "iae_ms": 2.16975488168198}}, {"scenario": {"payload0": 1.2, "t_load": 4.0, "dm": 11.0, "c0": 3.5, "F0": 2.0, "t_fric": 7.0, "fric_gain": 5.0, "delay": 6, "noise": 0.006, "T": 12.0, "seed": 3}, "ctrl": "adaptive", "expected": {"max_deviation_m": 0.7547817053580321, "settling_time_s": 5.710000000000001, "settled": true, "control_energy_kN2s": 0.2350248077818673, "step_overshoot_m": 0.07555317278466955, "final_error_m": 0.005155713014521002, "iae_ms": 1.1971782729946825}}, {"scenario": {"payload0": 3.9, "t_load": 4.0, "dm": -1.0, "c0": 3.5, "F0": 2.0, "t_fric": 7.0, "fric_gain": 1.0, "delay": 0, "noise": 0.0, "T": 12.0, "seed": 1}, "ctrl": "pid", "expected": {"max_deviation_m": 0.015467357512746749, "settling_time_s": 1.38, "settled": true, "control_energy_kN2s": 0.03196562041711937, "step_overshoot_m": 0.0723706591534844, "final_error_m": 8.959023124899268e-05, "iae_ms": 0.19405905613476956}}, {"scenario": {"payload0": 3.9, "t_load": 4.0, "dm": -1.0, "c0": 3.5, "F0": 2.0, "t_fric": 7.0, "fric_gain": 1.0, "delay": 0, "noise": 0.0, "T": 12.0, "seed": 1}, "ctrl": "adaptive", "expected": {"max_deviation_m": 0.03149865678202102, "settling_time_s": 0.8499999999999996, "settled": true, "control_energy_kN2s": 0.032041379768356255, "step_overshoot_m": 0.05414471009417465, "final_error_m": 0.0011271943141129004, "iae_ms": 0.22001462294797086}}]}