Search for: All records

Autonomous surgical robots are a promising solution to the increasing demand for surgery amid a shortage of surgeons. Recent work has proposed learning-based approaches for the autonomous manipulation of soft tissue. However, due to variability in tissue geometries and stiffnesses, these methods do not always perform optimally, especially in out-of-distribution settings. We propose, develop, and test the first application of uncertainty quantification to learned surgical soft-tissue manipulation policies as an early identification system for task failures. We analyze two different methods of uncertainty quantification, deep ensembles and Monte Carlo dropout, and find that deep ensembles provide a stronger signal of future task success or failure. We validate our approach using the physical daVinci Research Kit (dVRK) surgical robot to perform physical soft-tissue manipulation. We show that we are able to successfully detect out-of-distribution states leading to task failure and request human intervention when necessary while still enabling autonomous manipulation when possible. Our learned tissue manipulation policy with uncertainty-based early failure detection achieves a zero-shot sim2real performance improvement of 47.5% over the prior state of the art in learned soft-tissue manipulation. We also show that our method generalizes well to new types of tissue as well as to a bimanual soft-tissue manipulation task. 

Tendon-driven continuum robot kinematic models are frequently computationally expensive, inaccurate due to unmodeled effects, or both. In particular, unmodeled effects produce uncertainties that arise during the robot’s operation that lead to variability in the resulting geometry. We propose a novel solution to these issues through the development of a Gaussian mixture kinematic model. We train a mixture density network to output a Gaussian mixture model representation of the robot geometry given the current tendon displacements. This model computes a probability distribution that is more representative of the true distribution of geometries at a given configuration than a model that outputs a single geometry, while also reducing the computation time. We demonstrate uses of this model through both a trajectory optimization method that explicitly reasons about the workspace uncertainty to minimize the probability of collision and an inverse kinematics method that maximizes the likelihood of occupying a desired geometry. 

Tendon-driven continuum robots have been gaining popularity in medical applications due to their ability to curve around complex anatomical structures, potentially reducing the invasiveness of surgery. However, accurate modeling is required to plan and control the movements of these flexible robots. Physics-based models have limitations due to unmodeled effects, leading to mismatches between model prediction and actual robot shape. Recently proposed learning-based methods have been shown to overcome some of these limitations but do not account for hysteresis, a significant source of error for these robots. To overcome these challenges, we propose a novel deep decoder neural network that predicts the complete shape of tendon-driven robots using point clouds as the shape representation, conditioned on prior configurations to account for hysteresis. We evaluate our method on a physical tendon-driven robot and show that our network model accurately predicts the robot's shape, significantly outperforming a state-of-the-art physics-based model and a learning-based model that does not account for hysteresis. 

Abstract Nest‐site fidelity is a common strategy in birds and is believed to be adaptive due to familiarity with local conditions. Returning to previously successful nest sites (i.e., the win‐stay lose‐switch strategy) may be beneficial when habitat quality is spatially variable and temporally predictable; however, changes in environmental conditions may constrain dispersal decisions despite previous reproductive success. We used long‐term (2000–2017) capture‐mark‐reencounter data and hierarchical models to examine fine‐scale nest‐site fidelity of emperor geese (Anser canagicus) on the Yukon–Kuskokwim Delta in Alaska. Our objectives were to quantify nest‐site dispersal distances, determine whether dispersal distance is affected by previous nest fate, spring timing, or major flooding events on the study area, and determine if nest‐site fidelity is adaptive in that it leads to higher nest survival. Consistent with the win‐stay lose‐switch strategy, expected dispersal distance for individuals that failed their nesting attempt in the previous year was greater (207.7 m, 95% HPDI: 151.1–272.7) than expected dispersal distance for individuals that nested successfully in the previous year (125.5 m, 95% HPDI: 107.1–144.9). Expected dispersal distance was slightly greater following years of major flooding events for individuals that nested successfully, although this pattern was not observed for individuals that failed their nesting attempt. We did not find evidence that expected dispersal distance was influenced by spring timing. Importantly, dispersal distance was positively related to daily survival probability of emperor goose nests for individuals that failed their previous nesting attempt, suggesting an adaptive benefit to the win‐stay lose‐switch strategy. Our results highlight the importance of previous experience and environmental variation for informing dispersal decisions of a long‐lived goose species. However, it is unclear if dispersal decisions based on previous experience will continue to be adaptive as variability in environmental conditions increases in northern breeding areas. 

Tendon-driven continuum robot kinematic models are frequently computationally expensive, inaccurate due to unmodeled effects, or both. In particular, unmodeled effects produce uncertainties that arise during the robot’s operation that lead to variability in the resulting geometry. We propose a novel solution to these issues through the development of a Gaussian mixture kinematic model. We train a mixture density network to output a Gaussian mixture model representation of the robot geometry given the current tendon displacements. This model computes a probability distribution that is more representative of the true distribution of geometries at a given configuration than a model that outputs a single geometry, while also reducing the computation time. We demonstrate one use of this model through a trajectory optimization method that explicitly reasons about the 

ObjectiveTo examine the impact of increased body mass index (BMI) on (1) tracheotomy timing and (2) short‐term surgical complications requiring a return to the operating room and 30‐day mortality utilizing data from the Multi‐Institutional Study on Tracheotomy (MIST). MethodsA retrospective analysis of patients from the MIST database who underwent surgical or percutaneous tracheotomy between 2013 and 2016 at eight institutions was completed. Unadjusted and adjusted logistic regression analyses were used to assess the impact of obesity on tracheotomy timing and complications. ResultsAmong the 3369 patients who underwent tracheotomy, 41.0% were obese and 21.6% were morbidly obese. BMI was associated with higher rates of prolonged intubation prior to tracheotomy accounting for comorbidities, indication for tracheotomy, institution, and type of tracheostomy (p = 0.001). Morbidly obese patients (BMI ≥35 kg/m²) experienced a longer duration of intubation compared with patients with a normal BMI (median days intubated [IQR 25%–75%]: 11.0 days [7–17 days] versus 9.0 days [5–14 days];p < 0.001) but did not have statistically higher rates of return to the operating room within 30 days (p = 0.12) or mortality (p = 0.90) on multivariable analysis. This same finding of prolonged intubation was not seen in overweight, nonobese patients when compared with normal BMI patients (median days intubated [IQR 25%–75%]: 10.0 days [6–15 days] versus 10.0 days [6–15 days];p = 0.36). ConclusionBMI was associated with increased duration of intubation prior to tracheotomy. Although morbidly obese patients had a longer duration of intubation, there were no differences in return to the operating room or mortality within 30 days. Level of Evidence3Laryngoscope, 134:4674–4681, 2024