cs.RONov 2, 2026

Clinician-AI Collaboration in Endovascular Neurointerventions: In-Vitro User Study of Copilot Assistance Modes for Robotic Navigation Across the Autonomy Spectrum

Authors: Nikola Fischer, Harry Robertshaw, S. M. Hadi Sadati, Xingyu Chen, Weiyuan Deng, Benjamin Jackson, Michael Adofo Kwakye, Sara Sciacca, 

Organizations: School of Biomedical Engineering & Imaging Sciences, King’s College London, UK · School of Engineering and Materials Science, Queen Mary University of London, UK · Department of Neuroradiology, King’s College Hospital NHS Foundation Trust, London, UK · The Walton Centre NHS Foundation Trust, Liverpool, UK · 

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Mechanical thrombectomy (MT) is an effective treatment for acute ischemic stroke, yet access remains limited by the availability of neurointerventional expertise. Telerobotic and AI-assisted systems may help address this challenge, but it remains unclear how clinicians interact with different levels of autonomy and how autonomous navigation systems perform in realistic environments. This work introduces Copilot Assistance Modes (CAMs), a framework for clinician-AI collaboration in robotic endovascular navigation spanning the autonomy spectrum from manual operation to full autonomy. In the first in-vitro user study comparing clinician-AI collaboration strategies, 11 experts and two students evaluated six navigation modes in a patient-specific vascular phantom. Objective performance metrics were combined with subjective assessments, including workload and usability. Autonomous performance was additionally evaluated both in silico and in vitro to investigate sim-to-real transfer. Hands-on manual navigation achieved the strongest overall performance (100% success). While fully autonomous navigation achieved 81% success in simulation, performance dropped to 23% in vitro. In contrast, several clinician-AI shared-autonomy modes achieved success rates of 92-97%, approximately four times higher than full autonomy and comparable to direct robotic control (95%), demonstrating mitigation of sim-to-real domain shift through clinician-AI collaboration. Among the evaluated AI-driven approaches, user-directed task delegation and rotational shared control demonstrated the most favorable balance between performance and clinician acceptance. These findings suggest that effective clinician-AI collaboration and allocation of control authority may currently be more important for reliable robotic neurointerventional navigation than increasing autonomy alone.

Figures & tables

Fig. 1: Clinician-AI Collaboration for Neurointerventions: Introducing the Copilot Assistance Modes (CAMs) including the conventional Manual (0) and Direct Robot Control (1) as well as the AI-driven modes (blue) AI Advice (2), AI Assistance (3), Full Copilot Assistance (4), and Fully Autonomous Navigation (5). The blue color gradient indicates increasing levels of autonomy, ranging from white (manual) over grey (teleoperation) to dark blue (highest autonomy).
Fig. 2: Experimental Setup: (a) Navigation path (white dashed line) from the descending aorta through the aortic arch and brachiocephalic artery (BCA) into the right common carotid artery (RCCA), bypassing the right subclavian artery (RSA), and proceeding to either the right internal carotid artery (RICA) or the right external carotid artery (RECA). (b) Control room from which the participants are teleoperating the robotic manipulator utilizing the User Control Input Device (UCID) [ 25 ] . (c) A red Copilot push-button functions as a dead-man switch, activating autonomous navigation. (d) Mock operating room with the robotic manipulator for catheter and guidewire [ 26 ] , navigating within the in-vitro phantom under fluoroscopy imaging.
ParameterValueParameterValue
LSTM size500Batch size32
MLP layers4Replay buffer10610^{6}
MLP hidden units900Exploration noise0.25
Learning rate2.2×10−42.2\times 10^{-4}Training steps10M
Discount factor γ\gamma0.99Warm-up steps500K
Target update rate τ\tau0.005Update ratio0.05
TABLE I: Key hyperparameters for soft actor-critic algorithm.
#CAMRotationTranslation
0Manual (baseline)Manual φMan\varphi_{\mathrm{Man}}Manual dMand_{\mathrm{Man}}
1Direct Robot ControlTeleoperated φInp\varphi_{\mathrm{Inp}}Teleoperated dInpd_{\mathrm{Inp}}
2AI AdviceTeleoperated φInp\varphi_{\mathrm{Inp}}Teleoperated dInpd_{\mathrm{Inp}}
3AI AssistanceSplitting Robotic DOF
3aAI TranslationTeleoperated φInp\varphi_{\mathrm{Inp}}AI agent dAId_{\mathrm{AI}}
3bAI RotationAI agent φAI\varphi_{\mathrm{AI}}Teleoperated dInpd_{\mathrm{Inp}}
TABLE II: Taxonomy of Copilot Assistance Modes (CAMs)
Fig. 3: Control Architectures for AI-Driven Modes: Collaborative control loops for Copilot Assistance Modes (CAMs) (2)–(5), with AI components highlighted in blue. The operator controls the robot via the User Controller Input Device (UCID), which manipulates the guidewire within the 3D-printed vascular phantom. Instrument tip coordinates are extracted from the fluoroscopic image stream and provided to the AI agent. Grey components in CAMs (3)–(5) denote identical components. The symbol “ ⊗\otimes ” represents the combination of inputs in AI Assistance (3), whereas “ ⊕\oplus ” denotes a mutually exclusive selection between inputs in Full Copilot Assistance (4).
Fig. 4: AI-Advice User Interface (UI) : (a) Participant navigating the guidewire via robotic manipulator with the User Controller Interface Device (UCID) from the control room. (b) AI Advice is provided by the UI next to the fluoroscopic image. Direction of motion is highlighted in green for both translation and rotation while the red needle in the dial gauges quantify the amplitude of motion. Note, that labels for the right subclavian artery (RSA), right common carotid artery (RCCA), right external carotid artery (RECA), and right internal carotid artery (RICA) were not shown on the UI to the operator during the experiments.
Fig. 5: Success Rates Comparison: Stacked bars show the proportions of successful, failed (red), and timed-out (black) runs for each of the Copilot Assistance Modes (CAMs) (TABLE II ). Percentages within the bars indicate success and failure rates, and values above the bars indicate the total number of runs (n). For each participant, CAMs (0), (1), and (4), i.e. 4a and 4b combined, were tested in 6 runs, whereas CAMs (2), (3a), and (3b) were each tested in 4 runs. Fully Autonomous Navigation (5) was evaluated without participant involvement in 200 autonomous runs (100 targeting the right internal carotid artery (RICA) and 100 targeting the right external carotid artery (RECA)). The blue color gradient indicates increasing levels of autonomy, ranging from white (manual) over grey (teleoperation) to dark blue (highest autonomy).
Fig. 6: Time-to-Target (T2T): Distribution of T2T for all successful runs of each Copilot Assistance Mode (CAM). n values denote the number of successful runs per condition. Significance levels are denoted based on Holm-adjusted p-values: ( ∗\ast ), ( 0.01≤pHolm<0.050.01\leq p_{\mathrm{Holm}}<0.05 ); ( ∗∗∗\ast\ast\ast ), ( pHolm<0.001p_{\mathrm{Holm}}<0.001 ). The blue color gradient indicates increasing levels of autonomy, ranging from white (manual) over grey (teleoperation) to dark blue (highest autonomy).
Fig. 7: Errors: Distribution of navigation errors per run (filled) and per minute (hatched) for all successful runs of each Copilot Assistance Mode (CAM). Navigation errors correspond to deviations into incorrect vessel branches (Section II-F ). n denotes the number of successful runs per CAM . Fully Autonomous Navigation (5) is absent because any navigation error resulted in an unsuccessful trial. Unlike the clinician-supervised modes, in this mode, the AI agent could not recover from these errors, leaving no successful trials to display. The blue color gradient indicates increasing levels of autonomy, ranging from white (manual) over grey (teleoperation) to dark blue (highest autonomy).
Fig. 8: AI activation and deactivation during Full Copilot–Operator-Defined Sections (4b): Heatmaps depict the spatial distribution of AI activations (left column) and deactivations (right column) during navigation to the right external carotid artery (RECA, top row) and right internal carotid artery (RICA, bottom row). Values represent the percentage of AI-assisted interactions per grid cell and are overlaid on fluoroscopic phantom images. Warmer colors indicate a higher frequency of AI activation or deactivation within the corresponding region.
Fig. 9: Subjective Workload: Heatmap of NASA-TLX ratings by subscale and condition. Values in each cell are mean ± SD (displayed on two lines) on a 1–5 scale, with the Performance subscale reverse-coded to align interpretation (higher score = higher perceived workload).
Fig. 10: System Usability: Distribution of System Usability Score (SUS) for all CAM (expert clinician participants only, N=11N=11 ): boxplots show mean, median, and IQR (whiskers: 1.5×IQR); points show individual participants, shaded bands indicate standard SUS grade ranges (A–F). The SUS ratings for each item ranged from 00 – 55 and were converted to a 00 – 100100 scale. The blue color gradient indicates increasing levels of autonomy, ranging from grey (teleoperation) to dark blue (highest autonomy).
#CAM *MedianMean ±\pm SDRangeGrade
4Full Copilot77.573.4 ±\pm 20.922.5–100.0B
1Direct Robot Control75.074.8 ±\pm 15.945.0–90.0B
3bAI Rotation72.573.6 ±\pm 14.447.5–97.5B
2AI Advice57.553.4 ±\pm 26.517.5–92.5C
3aAI Translation50.053.6 ±\pm 21.320.0–90.0C
TABLE III: System Usability Scale (SUS) scores by Copilot Assistance Modes (CAMs)
#CAM *Median (IQR)Mean ±\pm SD1st ChoiceLast Choice
0Manual1 [1.0–1.5]1.64 ±\pm 1.298/110/11
1Direct Robot Control3 [2.0–3.5]3.00 ±\pm 1.180/110/11
4Full Copilot3 [2.0–4.5]3.09 ±\pm 1.702/111/11
2AI Advice4 [3.0–6.0]4.18 ±\pm 1.781/114/11
3bAI Rotation4 [4.0–4.5]4.09 ±\pm 0.700/110/11
3aAI Translation6 [4.5–6.0]5.00 ±\pm 1.410/116/11
TABLE IV: Ranking-based evaluation of Copilot Assistance Modes (CAMs)
More in this topic

Human-AI Collaboration

Human-AI collaboration studies how people and AI systems divide, coordinate, and adapt work to support decisions, creativity, and learning.

See all 458 papers

Explore similar work

CardsList
Mar 30, 2026

A Position Statement on Endovascular Models and Effectiveness Metrics for Mechanical Thrombectomy Navigation, on behalf of the Stakeholder Taskforce for AI-assisted Robotic Thrombectomy (START)

While we are making progress in overcoming infectious diseases and cancer; one of the major medical challenges of the mid-21st century will be the rising prevalence of stroke. Large vessels occlusions are especially debilitating, yet effective treatment (needed within hours to achieve best outcomes) remains limited due to geography. One solution for improving timely access to mechanical thrombectomy in geographically diverse populations is the deployment of robotic surgical systems. Artificial intelligence (AI) assistance may enable the upskilling of operators in this emerging therapeutic delivery approach. Our aim was to establish consensus frameworks for developing and validating AI-assisted robots for thrombectomy. Objectives included standardizing effectiveness metrics and defining reference testbeds across in silico, in vitro, ex vivo, and in vivo environments. To achieve this, we convened experts in neurointervention, robotics, data science, health economics, policy, statistics, and patient advocacy. Consensus was built through an incubator day, a Delphi process, and a final Position Statement. We identified that the four essential testbed environments each had distinct validation roles. Realism requirements vary: simpler testbeds should include realistic vessel anatomy compatible with guidewire and catheter use, while standard testbeds should incorporate deformable vessels. More advanced testbeds should include blood flow, pulsatility, and disease features. There are two macro-classes of effectiveness metrics: one for in silico, in vitro, and ex vivo stages focusing on technical navigation, and another for in vivo stages, focused on clinical outcomes. Patient safety is central to this technology's development. One requisite patient safety task needed now is to correlate in vitro measurements to in vivo complications.

Harry Robertshaw, Anna Barnes, Phil Blakelock +20

School of Biomedical Engineering & Imaging Sciences, Kings College London, London, UK · King’s Technology Evaluation Centre (KiTEC), Kings College London, London, UK · Patient & Public Involvement, Kings College London, London, UK +18

Robotic Endovascular InterventionSurgical RoboticsRobotics
AskPDF
Apr 22, 2026

Toward Safe Autonomous Robotic Endovascular Interventions using World Models

Autonomous mechanical thrombectomy (MT) presents substantial challenges due to highly variable vascular geometries and the requirements for accurate, real-time control. While reinforcement learning (RL) has emerged as a promising paradigm for the automation of endovascular navigation, existing approaches often show limited robustness when faced with diverse patient anatomies or extended navigation horizons. In this work, we investigate a world-model-based framework for autonomous endovascular navigation built on TD-MPC2, a model-based RL method that integrates planning and learned dynamics. We evaluate a TD-MPC2 agent trained on multiple navigation tasks across hold out patient-specific vasculatures and benchmark its performance against the state-of-the-art Soft Actor-Critic (SAC) algorithm agent. Both approaches are further validated in vitro using patient-specific vascular phantoms under fluoroscopic guidance. In simulation, TD-MPC2 demonstrates a significantly higher mean success rate than SAC (58% vs. 36%, p < 0.001), and mean tip contact forces of 0.15 N, well below the proposed 1.5 N vessel rupture threshold. Mean success rates for TD-MPC2 (68%) were comparable to SAC (60%) in vitro, but TD-MPC2 achieved superior path ratios (p = 0.017) at the cost of longer procedure times (p < 0.001). Together, these results provide the first demonstration of autonomous MT navigation validated across both hold out in silico data and fluoroscopy-guided in vitro experiments, highlighting the promise of world models for safe and generalizable AI-assisted endovascular interventions.

Harry Robertshaw, Nikola Fischer, Han-Ru Wu +6

Surgical & Interventional Engineering, School of Biomedical Engineering & Imaging Sciences, King’s College London, UK · Department of Radiology, National Taiwan University Hospital, Taiwan · Department of Neuroradiology, King’s College Hospital, UK

World Models for RoboticsRobotic Endovascular InterventionSurgical RoboticsModel-Based RL
AskPDF
Feb 20, 2026

Toward AI Autonomous Navigation for Mechanical Thrombectomy using Hierarchical Modular Multi-agent Reinforcement Learning (HM-MARL)

Mechanical thrombectomy (MT) is typically the optimal treatment for acute ischemic stroke involving large vessel occlusions, but access is limited due to geographic and logistical barriers. Reinforcement learning (RL) shows promise in autonomous endovascular navigation, but generalization across 'long' navigation tasks remains challenging. We propose a Hierarchical Modular Multi-Agent Reinforcement Learning (HM-MARL) framework for autonomous two-device navigation in vitro, enabling efficient and generalizable navigation. HM-MARL was developed to autonomously navigate a guide catheter and guidewire from the femoral artery to the internal carotid artery (ICA). A modular multi-agent approach was used to decompose the complex navigation task into specialized subtasks, each trained using Soft Actor-Critic RL. The framework was validated in both in silico and in vitro testbeds to assess generalization and real-world feasibility. In silico, a single-vasculature model achieved 92-100% success rates on individual anatomies, while a multi-vasculature model achieved 56-80% across multiple patient anatomies. In vitro, both HM-MARL models successfully navigated 100% of trials from the femoral artery to the right common carotid artery and 80% to the right ICA but failed on the left-side vessel superhuman challenge due to the anatomy and catheter type used in navigation. This study presents the first demonstration of in vitro autonomous navigation in MT vasculature. While HM-MARL enables generalization across anatomies, the simulation-to-real transition introduces challenges. Future work will refine RL strategies using world models and validate performance on unseen in vitro data, advancing autonomous MT towards clinical translation.

Harry Robertshaw, Nikola Fischer, Lennart Karstensen +6

School of Biomedical Engineering & Imaging Sciences, Kings College London, London, UK · School of Engineering & Materials Science, Queen Mary University of London, London, UK

Robotic Endovascular InterventionMulti-Agent Reinforcement LearningHierarchical RL
AskPDF