Interaction Dynamics Modeling and Predictive Control for Safe Steerable Catheter--Tissue Interaction
arXiv:2607.20939
2026
Dynamics
2 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper offers a transferable control pattern rather than a catheter-specific design: cancel only reliable nominal physics, represent unmodeled interaction effects with an augmented disturbance state, and enforce safety through explicit finite-horizon constraints. Its key quantitative mechanism is offset-free regulation from a disturbance-augmented state model, while hard force constraints prevent the disturbance-rejection controller from trading tracking error for unsafe contact. This can be transferred to neural state-space models and learned controllers by adding a slowly varying latent disturbance channel and placing a predictive safety layer around the network. The most useful experiments should test the predicted difference between unconstrained offset correction and constrained prediction near a learned safety boundary.
Ideas from this paper
✗ Mechanism failed
2026
Wrap a neural policy or neural dynamics model in a short-horizon predictive optimizer that enforces explicit bounds on a learned interaction variable before applying the next action. This separates disturbance rejection and tracking from safety: the network may propose aggressive corrections, but the optimizer projects them onto actions whose predicted force, state, and actuator trajectories remain feasible.
Useful7/10
Difficulty6/10
Novelty5/10
✗ Failed on benchmark
2026
Augment a neural recurrent or state-space model with an explicit slowly varying disturbance state that absorbs contact effects, friction, hysteresis, actuator mismatch, and other systematic residuals. The network predicts nominal dynamics, while the disturbance channel provides offset-free correction without forcing the main model to memorize every operating-condition-dependent bias.
Useful7/10
Difficulty5/10
Novelty6/10