Co-Learning Port-Hamiltonian Systems and Optimal Energy-Shaping Control

2026-04-30 · Source: stat.ML updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

Researchers from Michigan State University and Meta Reality Labs have developed a physics-informed learning framework for energy-shaping control of port-Hamiltonian (pH) systems using trajectory data. The approach, supported by NSF Award CMMI-1940950, co-learns a pH system model and an optimal energy-balancing passivity-based controller (EB-PBC) through alternating optimization with policy-aware data collection. Both the system model and controller are parameterized by neural networks that embed pH dynamics and EB-PBC structure, ensuring interpretability and stability. A key feature is a dissipation regularization that enforces strict energy decay during training, enhancing robustness to sim-to-real gaps. The framework was validated on state-regulation and swing-up tasks for planar and torsional pendulum systems, demonstrating improved control effort efficiency compared to traditional methods.

Key takeaway

For research scientists developing control systems for complex physical dynamics, this framework offers a robust method to synthesize high-performance controllers. You should consider integrating physics-informed neural networks and alternating optimization to ensure both model accuracy and provable stability, especially when facing model uncertainty or sim-to-real transfer challenges. The dissipation regularization is critical for practical deployment.

Key insights

Co-learning pH system models and optimal EB-PBC policies via alternating optimization ensures stability and robustness in control.

Principles

• Embed physical structure into neural networks for interpretability.
• Alternating optimization improves model fidelity in policy-relevant regions.
• Dissipation regularization enhances robustness to model mismatch.

Method

The framework alternates between refining the pH system model using mixed step-excited and policy-excited trajectory data, and re-optimizing the EB-PBC controller on the updated model, with neural networks parameterizing both components.

In practice

• Apply to robotics and aerospace for high-performance control.
• Use for systems with unknown or partially known dynamics.
• Implement dissipation regularization for sim-to-real robustness.

Topics

• Port-Hamiltonian Systems
• Energy-Balancing Passivity-Based Control
• Physics-Informed Neural Networks
• Alternating Optimization
• Sim-to-Real Robustness

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Dual-Network PINNs for Optimal Control: A Reproducible Benchmark on the Mass-Spring-Damper System — A dual-network PINN formulation achieves high accuracy in optimal control by exactly enforcing boundary conditions, offering a clear, reproducible alternative to classical methods.
• Structure-Oriented Randomized Neural Networks for Poisson-Nernst-Planck and Poisson-Nernst-Planck-Navier-Stokes Systems — SO-RaNN uses randomized neural networks with structure-preserving corrections to solve complex PNP and PNP-NS systems.
• Gradient descent at the Edge of Stability: free energy model and kinetic description of the two-layer network — A new model tracks average trajectory and oscillation covariance, revealing an effective free energy for Edge of Stability dynamics.
• Learning Over-Relaxation Policies for ADMM with Convergence Guarantees — Learning ADMM relaxation parameters online improves performance without costly matrix refactorizations.
• Meta-Learning for Rapid Adaptation in Reference Tracking of Uncertain Nonlinear Systems — A meta-learning control framework enables rapid adaptation for uncertain nonlinear systems using limited target data and offline source system knowledge.
• Self-orthogonalizing attractor neural networks emerging from the free energy principle — Attractor networks emerge from the Free Energy Principle, yielding self-organizing, orthogonal representations and robust learning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.