*Important notice: This news reports on an unedited version of an accepted paper and is awaiting final editing. Therefore, the paper should not be regarded as conclusive or treated as established information.
Researchers have investigated how task complexity and motion scaling ratio influence how humans learn to teleoperate a surgical robot. Using a ring-through-rail task on the da Vinci Research Kit, the authors compared human learning with a control-theory model and found that a simple algorithm can capture key features of human motor learning.
Study: Human learning dynamics during teleoperation of surgical robots. Image Credit: Terelyuk/Shutterstock.com
Navigating the Surgical Robot Learning Curve
Robot-assisted minimally invasive surgery (RAMIS) has become integral to modern clinical practice, offering enhanced precision, tremor reduction, and motion scaling. However, mastering these systems presents unique challenges, where surgeons must coordinate eye-hand-foot movements at a remote console, relying on video-mediated three-dimensional (3D) imagery without tactile feedback.
These altered visual-motor mappings create a steep learning curve. While simulation-based training curricula are increasingly used, researchers have not systematically analyzed how task complexity influences skill retention and transfer in RAMIS.
Skill acquisition also depends on a surgeon's ability to adapt to the robotic interface, constructing internal models of how the robot responds to control inputs. This study disentangled task complexity and motion scaling effects by comparing human learning with a control-theory model performing identical tasks.
Click here to download a free PDF copy of this page
Measuring Human Learning in Teleoperation
Participants performed a ring-through-rail teleoperation task using a da Vinci Research Kit system. They controlled robotic grippers from the surgeon console, guiding a ring along a fixed 3D rail while minimizing contact.
Visual feedback came from a stereo endoscopic camera. This task is a validated exercise from the Fundamentals of Robotic Surgery curriculum and served as an abstraction of surgical scenarios, allowing systematic manipulation of geometric features such as curvature and elevation changes.
We derived a composite performance score from multiple behavioral metrics. For each metric, we normalized values so higher scores consistently reflected better performance. The composite score combined these values using weights from a principal component analysis, yielding a measure from zero to 100.
To evaluate learning, we compared performance in the first training block with the retention block using paired t-tests and effect sizes. Linear mixed-effects models assessed block-wise improvement over training.
Retention was defined as the relative change between retention and the first training block. Transfer was defined as the relative change between the transfer block and pre-training baseline. We modeled these rates with complexity and ratio as fixed effects.
Beyond performance scores, we modeled the human teleoperator as an optimal feedback controller that maintained an internal estimate of the motion scaling ratio and updated it through training.
The model was simulated under the same experimental conditions as human participants. For comparison, we also implemented simpler controllers, but the optimal feedback model showed the strongest similarity to human correction dynamics.
Human and Model Learning Patterns
Human performance improved significantly across training. Translational path length, trial duration, collision frequency, collision duration, and number of drops all decreased from the first training block to retention, with medium-to-large effect sizes.
Rotational path length, however, did not reliably decrease, suggesting participants stabilized end-effector movement while continuing to explore rotational strategies.
A composite performance score derived from principal component analysis increased significantly across training blocks. Subjective workload, measured by the National Aeronautics and Space Administration Task Load Index (NASA-TLX), decreased consistently, and lower workload was reliably associated with higher performance scores.
The linear-quadratic-Gaussian (LQG) controller qualitatively reproduced several trajectory features observed in human performance. When simulated under matched conditions, it improved over training similarly to humans.
For retention and transfer, task complexity showed significant effects on both outcomes, with large effect sizes. Transfer performance decreased with increasing complexity, while retention improvement showed the opposite pattern, strongest at the highest complexity level.
This dissociation suggests that higher complexity rails provided greater opportunity for within-condition improvement but reduced generalization to new conditions. Motion scaling ratio effects were small, and human participants showed no clear asymmetric transfer pattern.
In LQG simulations, transfer revealed significant effects of both complexity and ratio on path length metrics, while contact-related outcomes were driven primarily by task complexity. This suggests geometric constraints dominated performance, while motion scaling primarily affected movement efficiency.
Understanding the Drivers of Teleoperation Learning
This study examined how task complexity and motion scaling influence learning during surgical robot teleoperation by having humans and an LQG controller perform identical tasks.
Both showed similar learning patterns, suggesting a control-theoretic model can capture key aspects of teleoperation skill acquisition, where operators update internal estimates while balancing accuracy and effort. Both were sensitive to task complexity, with geometric constraints affecting performance more strongly than scaling ratio.
However, the model remains a simplified baseline and lacks cognitive factors like fatigue, attention, and strategic exploration. Limitations include non-surgeon participants, unimanual task design, and simplified kinematics. Despite this, the LQG framework offers a useful baseline for understanding teleoperation learning.
Toward Smarter Surgical Robot Training
This study demonstrates that a simple control-theory model can capture fundamental features of how humans learn to teleoperate surgical robots. By comparing human participants with an LQG controller performing identical ring-through-rail tasks, the authors showed that both learners improved similarly with practice and responded comparably to task complexity.
Geometric constraints emerged as the dominant factor influencing retention and transfer, while motion scaling primarily affected movement efficiency. Although the model remains simplified, it provides a valuable baseline for understanding teleoperation learning. These findings offer a mechanistic framework that could inform surgical training protocols and robotic interface design.
Journal Reference
Huang, Y., Cai, Y., Li, M., Chen, Y., & Wilson, R. C. (2026). Human learning dynamics during teleoperation of surgical robots. Npj Science of Learning. DOI:10.1038/s41539-026-00442-6, https://www.nature.com/articles/s41539-026-00442-6
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.