Neural networks ensure safety while modeling complex system dynamics
Safe-by-Design Learning via Energy-based Neural Networks
Machine LearningArtificial Intelligence
Summary
Controlling machines safely is crucial but hard when their behavior is complicated. The authors designed a new kind of neural network that naturally keeps a system within safe limits, without requiring expensive safety checks afterward. They combined ideas from energy-based networks and physics-inspired models to build this system. Their tests showed it works well on challenging problems like controlling small drones.
What this means in practice
- •For autonomous vehicle engineers: Design autonomous systems that safely operate within certified boundaries, reducing the risk of failures in dynamic environments.
- •For robotics control teams: Develop controllers for drones and robots that guarantee safety by construction, enabling more reliable operations under disturbances.
Authors
Simone Betteti, Morteza Lahijanian, Luca Laurenti
Abstract
Learning neural-network models of dynamical systems with safety guarantees is a fundamental requirement for their deployment in safety-critical settings. Safety is commonly established by proving the invariance of a desired subset in state-space, ensuring that every trajectory initialized in this subset remains confined to it for all time under admissible inputs. Existing frameworks, however, either rely on computationally expensive post-hoc verification or employ safety-enforcing mechanisms without formal correctness guarantees. In this paper, we introduce a novel neural architecture grounded in energy-based modern Hopfield networks to guarantee safety-by-design while retaining sufficient expressiveness to model complex nonlinear dynamics. Specifically, we integrate modern Hopfield networks with a port-Hamiltonian neural ODE, enabling by design the construction of barrier functions yielding explicit admissible-input sets and quantitative robustness radii. Across several benchmarks, including an 12-dimensional nanodrone model, our framework achieves state-of-the-art performance while producing certified invariant sets that are more robust to external solicitations than comparable existing approaches.