August 06, 2026
More on why Sentante’s physical-instrument interface determines whether captured demonstrations reproduce expert clinical skill.
Sentante today published the second in its Physical AI series, setting out why the operator interface is among the most consequential design decisions in any endovascular robotic platform.
Sentante also sets out why the 1:1 motion-mimicking physical-instrument interface is the data-quality foundation that determines whether imitation-learning policies trained on captured procedures reproduce expert clinical skill, or a filtered approximation of it.
A reverse-framing question:
If you had to digitise an expert interventionalist’s career – fifteen years of feel, judgement, and micro-adjustments – so a machine could learn from it, how would you do it?
You would not ask the expert to translate everything through a joystick before the device responds. You would capture exactly what they do, in their natural mode, while they do it. The expertise stays in the operator’s hands. The recording captures their motor program as it actually runs. That is the design choice the Sentante interface makes.
The hidden interface tax:
An interventionalist accumulates expertise over a career – the micro-adjustments, the torque compensations, the way the hand reads the tension in the instrument, pushability through a long compliant column. These are decades of pattern recognition encoded in the operator’s motor program. They are also the demonstrations on which a Physical AI policy is trained. The question is whether the interface preserves them.
Every joystick, mouse, or screen-based controller interface surveyed in the endovascular robotic field in 2026 forces the operator to translate their natural motor program through an artificial input layer before the device responds. Two consequences follow. The captured demonstration records expert joystick operation, not expert catheter operation – the post-translation signal is recorded, the pre-translation expertise is not. And the demonstrator pool is bound to physicians who have completed a months-long learning curve on that specific joystick-controlled robot.
The 1:1 motion-mimicking architecture:
Sentante’s interface is mechanically and behaviourally different. The operator manipulates real catheters and guidewires as they would in manual intervention. The motion-mimicking architecture records the operator’s natural motor program directly, without translation through an artificial input modality. The captured behaviour carries the actual clinical skill – the same micro-adjustments, the same feel for pushability, the same torque compensation an experienced operator has built over a career.
The implication for imitation learning is structural. The data-scaling law for imitation learning (Lin et al., ICLR 2025) reports that policy generalisation depends on demonstrator and environment diversity more than on raw demonstration count. A 1:1 motion-mimicking interface produces two structural advantages in that paradigm. The captured behaviour is the expert’s natural motor program, recorded without filtering through an artificial interface, so the training signal carries the actual clinical skill. And the demonstrator pool collapses from “physicians trained on this specific robot” to “every existing expert interventionalist.”
The soft-tissue robotics precedent – why motion-mimicking became the standard:
Surgical robotics has been here before. When da Vinci entered minimally invasive surgery in 1999, the design choice that made it adoptable was not the camera or the wristed instruments – it was the master console. Surgeons sit at handles that mimic the motion of the instrument tips one-to-one, with motion scaling and tremor filtration on top. The master controller behaves as a natural extension of the surgeon’s hand. Skills built over a manual surgical career transfer to the robotic console without re-learning.
Every cleared soft-tissue surgical robot that followed – CMR Surgical’s Versius, Medtronic’s Hugo, Asensus – adopted the same motion-mimicking paradigm. None ships with a joystick. The reason is the same one Sentante applies to endovascular: when the existing manual motor program is the substrate from which the operator builds clinical expertise, the interface that preserves it is the interface that wins clinical adoption. The 2024 addition of force feedback in da Vinci 5 completed this logic in soft tissue.
Sentante applies the same logic from day one in endovascular, and applies it to the catheter and guidewire motor program that interventionalists have spent careers building.
Why this matters for clinical adoption:
The same design choice determines clinical adoption velocity. A platform that requires a multi-month learning curve before a physician can operate competently runs into the same training-bandwidth bottleneck that has constrained the adoption of every prior endovascular robotic platform. A platform on which an existing expert interventionist is competent from day one removes that bottleneck – and accumulates training data from day one.
Dr. Tomas Baltrunas, co-founder and Chief Medical Officer of Sentante, commented: “What I learned to do with my hands over fifteen years of intervention is not something I want to translate through a joystick mapping before the device responds. The motion-mimicking interface lets me operate as I always have, with the platform recording what I do as I do it. The expertise stays in the operator’s hands, and the data stays in the system.”
Edvardas Satkauskas, co-founder and CEO of Sentante, added: “The interface is where the data quality is made or lost. Every cath-lab procedure performed on a Sentante platform is a training-grade demonstration in its natural form. That is what the imitation-learning literature says you need for Physical AI, and that is what no joystick-controlled platform can supply.”
About Sentante:
Sentante is a medical robotics company founded in 2017 building a haptic, device-agnostic endovascular platform that enables clinicians to perform complex vascular procedures remotely with full tactile feedback. Designed to integrate with existing cath-lab infrastructure, Sentante aims to expand access, improve clinician safety, and elevate procedural consistency across peripheral vascular, neurovascular and cardiovascular applications. Sentante is a member of the NVIDIA Inception programme.
