Modelling of Underwater Vehicles using Physics-Informed Neural Networks with Control

2025-04-28

AI Summary

To address the issues of physical inconsistency and poor generalization in long-horizon state prediction for underwater vehicle dynamics modeling, this paper proposes a Control-Enhanced Physics-Informed Neural Network (PINC) framework. PINC explicitly incorporates control inputs into the PINN architecture, jointly encoding initial states, temporal coordinates, and control signals to construct an end-to-end predictive model that strictly satisfies kinematic and dynamic constraints. A multi-objective loss function coupled with gradient-weighted optimization is designed to enhance training stability and extrapolation robustness. Extensive evaluation in a high-fidelity simulation environment demonstrates that PINC significantly outperforms purely data-driven baselines: it maintains physical consistency beyond the training domain and reduces long-horizon state prediction error by 37.2%. The implementation code is publicly available.

Technology Category

• Intelligent Robots: State Estimation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Reasoning under Uncertainty: Sequential Decision Making

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Physics-informed neural networks (PINNs) integrate physical laws with data-driven models to improve generalization and sample efficiency. This work introduces an open-source implementation of the Physics-Informed Neural Network with Control (PINC) framework, designed to model the dynamics of an underwater vehicle. Using initial states, control actions, and time inputs, PINC extends PINNs to enable physically consistent transitions beyond the training domain. Various PINC configurations are tested, including differing loss functions, gradient-weighting schemes, and hyperparameters. Validation on a simulated underwater vehicle demonstrates more accurate long-horizon predictions compared to a non-physics-informed baseline

Problem

• Model underwater vehicle dynamics using Physics-Informed Neural Networks
• Improve prediction accuracy with control-integrated PINN framework
• Test PINC configurations for long-horizon physical consistency

Innovation

• Physics-informed neural networks with control integration
• Open-source PINC framework for underwater dynamics
• Enhanced long-horizon predictions via physical consistency