Answers are grounded in this paper and cite their sources.
Nonlinear dynamics and deformations of soft robots pose challenges to conventional robot modeling and control. Physics-informed control integrates systems physics in learning to achieve physical consistency and maintain adaptability. The goal of this review is to categorize and summarize physics-informed control strategies for soft robots and correlate the integration of physics in control performance and application relevance. Five categories are examined: model–data integration, physics-informed neural networks (PINNs), Lagrangian neural networks (LNNs), physics-informed reinforcement learning, and physics-informed Koopman operators. The literature is compared by incorporated physics, control objectives, performance, data and computational requirements, and experimental validation. The relevance of their applications is further explored in medical and assistive, industrial, agricultural, and underwater applications, taking into account domain-specific requirements and the maturity of the evidence available. The synthesis identifies the trade-offs between model fidelity, adaptability, and computational efficiency and highlights the key challenges for generalization, observability, model–actuation consistency, closed-loop stability, and real-world validation. These findings distinguish demonstrated control capabilities from application readiness and identify priorities for practically applicable soft-robot control.
No takes yet. Share an insight, caveat, or question.
Kalsoom et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: