Transverse feedback linearization (TFL) achieves path following by exactly linearizing the error dynamics, but its controller is undefined on a singular set, such as zero speed for a unicycle or zero thrust for a quadrotor. The constrained region of attraction, the set of initial states from which the closed loop converges to the path without violating state or regularity constraints, determines where the controller can be safely activated. Exploiting the exact linear error dynamics, we show that this region is the largest forward invariant subset of the admissible set. We then decompose its state-constrained Zubov value function into an analytically known quadratic cost-to-go and a safety value function. We learn only the latter, by fitted value iteration without infinite-horizon trajectory labels. The resulting estimate combines an inner layer verified by interval branch and bound with a learned outer layer tested on samples. On a unicycle following a circle and a sinusoid and on a planar quadrotor, it covers on average 72.0%, 83.4% and 19.1% of the simulated region of attraction. In comparison, the quadratic cost-to-go and two physics-informed neural network (PINN) solvers of Zubov's equation cover at most 11.3%, 4.0% and 0.0% on average. Code and animations are available at https://gradslab.github.io/roa_tfl
Figures & tables
Case
Share of box
Inner
Quadratic
Zubov PINN
Dirichlet- Zubov PINN
No decomp.
Ours
Exact v
FP
Circle
84.8
6.4
8.3
6.4±0.0
11.3±3.9
11.8±3.2
72.0±1.6
85.4
0.00
Sinusoid
95.0
1.1
2.1
4.0±0.6
1.1±0.0
31.4±20.9
83.4±0.8
88.1
0.00
Quadrotor
56.2
0.0
0.0
0.0±0.0
0.0±0.0
0.0±0.0
19.1±4.0
46.8
0.00
Table 1: Coverage (%) of the simulated region of attraction on the TFL cases, mean ± std over four seeds. Exact v : exact value function cut at the level of the candidate. FP: largest share of reference states inside an estimate and outside that region.
Figure 1: Slices of the three cases, the unicycle at heading north and speed 1 and the quadrotor at the orbit velocity, level tilt and hover thrust. The estimate is the union over four seeds.
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
Case
γ
Sz−1
Circle
0.0444
1.550, 1.396, 1.011
Sinusoid
0.0608
1.001, 0.625, 1.000
Quadrotor
0.0288
1.546, 2.439, 5.953, 0.703, 1.034
Appendix
Table 2: Constants computed before training: the cost scale γ of the long-step setting and the diagonal of Sz−1 .
System
Share, long
Share, short
Kept
Coverage, long
Coverage, short
Coverage, kept, fine-tuned
Circle
61.1
59.3
long
72.0
69.8
74.0
Sinusoid
78.4
54.3
long
82.5
57.2
83.7
Quadrotor
14.0
5.3
long
24.9
9.4
22.9
Van der Pol, μ=1
30.3
38.8
short
77.3
98.9
98.9
Van der Pol, μ=3
15.4
21.3
short
54.1
74.9
89.2
Pendulum
30.2
27.4
long
93.5
84.6
94.2
Appendix
Table 3: Step setting of Algorithm 4 on seed 0. Share: share of box samples in the estimate (%), before fine-tuning, which decides the setting without reference labels. Coverage (%) of the simulated region of attraction is shown for comparison, before fine-tuning for both settings and after fine-tuning for the kept one.
Case
cI
c1
I (% box)
Cells
Time (s)
Circle
0.0350
0.0344
5.40
944395
588
Sinusoid
0.0170
0.0168
1.01
1047557
110
Quadrotor
0.0002
0.0002
0.00
21270981
3626
Appendix
Table 4: Verified inner layers of the TFL cases, with the cells and time of the final verification.
Value function and candidate
Value function
bi=0.1⋅ median di , h=∑irelu(bi/di−1)2 , γ of the step setting, with q90 the 90th percentile of qex over admissible samples
Shaping cost
κ=0.3 if the inner layer holds at least 1% of 2⋅105 box samples, and κ=0 otherwise. The function ρ is a smooth step from γqex=cI/2 to cI . The share on the sinusoid is 1.01% , close to this threshold
Network
Nθ : 64-64-64 tanh. Features: normalized state (angles as sine and cosine), tanh(ζ/3) , tanh(21ln(di/mediandi)) for the constraints on ΔD and ∥∇ϖ∥−1 . Softplus sharpness β=10 . The constant q0 is the first percentile of qex on 2⋅105 collocation points
Initialization
The network starts close to Φθ=ψ , which lies below the exact solution
Training
Collocation
4⋅105 Sobol box samples, 105 near the faces, 105 next to the singular set (admissible ones kept)
Appendix
Table 5: Configuration of every case: value function, candidate and training. Step lengths are in normalized time.
Level selection and fine-tuning
Level selection
Horizons H∈{1,2,4,8} , RK4 substep 0.025 , states with ∥ζ∥≤10−3 , which lie in I , exempt from the decrease. Snapshots every 5 epochs and at epoch 0. A test on a number of box samples adds half as many near the faces, a fifth as many next to the singular set and half as many on the faces, where states with outward field count as counterexamples. Screening: 5⋅104 box samples, 3 rounds. Retest of the top 3: 105 box samples, 5 rounds. Ranking by share of 5⋅104 fixed box samples. Final test: 106 i.i.d. uniform box samples
Fine-tuning
4 rounds, each on the counterexamples with c1≤vθ<min{c2+0.05,0.999} for a level c2 estimated again in that round, pool of 106 box samples, 5⋅105 near the faces and 2⋅105 next to the singular set, 8 Gaussian neighbours per counterexample, 300 Adam steps (learning rate 5⋅10−5 , batch 4096), weight 100 on counterexamples whose step stays in Ωad , weight 10 on those that leave it, decrease margin 0.2, counterexample sets capped at 5⋅104 , anchor on 2⋅105 box samples, exponential loss weight with Vmax=3 on all systems
Reference simulation
Reference
5⋅104 Sobol states, RK4 step 0.01, horizon 30 (unicycle) and 40 (quadrotor). Converged: held for one time unit, ∥ξ∥<10−3 and ∣η2−1∣<0.05 (unicycle), ∥ζ∥<10−3 (quadrotor). States that have neither converged nor failed at the horizon count as outside
Appendix
Table 6: Configuration of every case: level selection, fine-tuning and reference simulation. Step lengths are in normalized time, except in the reference simulation.
Case
Seed
c2 before
Cov. before
Round
c2 sel.
c2 final
Annulus
Cov.
FP
Circle
0
0.8695
71.99
4
0.9700
0.9700
574520
74.04
0.000
Circle
1
0.8570
69.16
4
0.9768
0.9768
544826
70.46
0.000
Circle
2
0.8539
69.82
3
0.9786
0.9739
570552
72.52
0.000
Circle
3
0.8358
69.23
4
0.9687
0.9687
548199
70.85
0.000
Sinusoid
0
0.9071
82.53
1
0.9770
0.9770
784823
83.75
0.000
Sinusoid
1
0.8668
81.64
1
0.9722
0.9700
790751
84.01
0.000
Appendix
Table 7: Per-seed results. Before: the trained network. Round: selected fine-tuning round (0 is the trained network). Sel. and Final: level after the retest and after the final test. Annulus: final-test samples between c1 and the selected level. FP as in Table 1 .
Case
Variant
c2
Coverage
FP
Circle
full method
0.869
72.0
0.000
Circle
PDE residual loss alone
0.195
10.5
0.000
Circle
safety-value baseline
0.743
11.1
0.002
Circle
no finite-step penalty, no counterexamples
0.768
63.4
0.000
Circle
no exponential loss weight
0.777
66.7
0.000
Circle
sum form Φ=ψ+Ψθ
0.995
9.9
0.000
Appendix
Table 8: Ablations on seed 0 before fine-tuning, with the same level selection and final test. FP: share (%) of reference states inside the estimate and outside the simulated region of attraction.
Department of Mechanical & Industrial Engineering at the New Jersey Institute of Technology, Newark, NJ 07102, USA · School of Engineering and Sustainable Development at the De Montfort University, Leicester, LE1 9BH, UK