← All AI/ML open problems

Control & robotics · AI safety & alignment · Verification & proof search / Open question

Robots with Provable Stability

A learned robot controller should remain dependable as its model changes and as separately designed components interact.

OpenEditorially scopedFramework needs definitionCompiled 2026-11-19

Seed catalog · Showing the bundled record.

Problem statement

For NN interconnected polynomial control subsystems on compact semialgebraic regions, use sample-and-hold neural controllers with period hh and coefficient uncertainty sets learned from data. Maintain local certificates satisfying V˙i≤−αiVi+∑j≠iγijVj+ϵi\dot V_i\leq-\alpha_i V_i+\sum_{j\ne i}\gamma_{ij}V_j+\epsilon_i and the corresponding small-gain condition during model/controller updates. Determine data and verification costs for robust practical stability and constraint satisfaction, with the disturbance, sampling and certificate-error margins explicit; hybrid resets require additional stated reset inequalities.

Assumptions and scope

  • Fix operating regions, switching/contact rules, disturbance bounds and sampled-data implementation; smooth continuous-time certificates do not automatically cover hybrid or discrete execution.
  • Use sound uncertainty sets tied to declared observations and specify interconnection or small-gain conditions; data fit alone does not bound every model error.
  • Define how certificates are revalidated or updated before deployment, and distinguish proof of a supplied certificate from a guarantee that training finds one.

What would settle it?

A compositional adaptive-certification theorem with explicit complexity and robust stability margins for a nontrivial system class, or a lower bound locating an unavoidable verification or information bottleneck.

Selected for a theoretical or computational target. A fitted scaling curve or improved benchmark score alone does not establish the stated learning-theoretic guarantee.

Status and evidence

Robust neural Lyapunov–barrier certificates for bounded perturbations and compositional neural Lyapunov verification already exist. The target is maintaining such guarantees during coupled learning and hybrid/sampled execution, not the first neural stability certificate. This is an editorial, source-motivated research target. The preliminary review through 2026-11-19 does not certify novelty or exclude an unlocated resolution.

Preliminary source check; not an expert-verified open status or an independent audit of proposed proofs. The compilation date is not a literature-review cutoff.

Sources and further reading

  1. Wang, Wu and Singh — Formal Synthesis of Certifiably Robust Neural Lyapunov-Barrier Certificates ↗partial theory · source checked 2026-11-19

    Primary abstract checked: sufficient Lipschitz conditions for bounded dynamics perturbations and robust certificate training. Robust neural safety/stability certificates already exist.

  2. Liu et al. — Compositionally Verifiable Vector Neural Lyapunov Functions for Interconnected Nonlinear Systems ↗partial theory · source checked 2026-11-19

    Primary abstract checked: compositional verification of neural Lyapunov functions exploits subsystem structure beyond monolithic SMT capability. Composition alone is not the proposed novelty.

  3. Luo, Sun and Kapoor — Sample-Efficient Safe Learning for Online Nonlinear Control with Control Barrier Functions ↗partial theory · source checked 2026-11-19

    Primary abstract checked: episodic optimistic control with stochastic barrier constraints, high-probability safety and regret analysis under specified assumptions.