Augmented Reality (AR) benefits telementoring by enhancing the communication between the mentee and the remote mentor with mentor authored graphical annotations that are directly integrated into the mentee’s view of the workspace. An important problem is conveying the workspace to the mentor effectively, such that they can provide adequate guidance. AR headsets now incorporate a frontfacing video camera, which can be used to acquire the workspace. However, simply providing to the mentor this video acquired from the mentee’s first-person view is inadequate. As the mentee moves their head, the mentor’s visualization of the workspace changes frequently, unexpectedly, and substantially. This paper presents a method for robust high-level stabilization of a mentee first-person video to provide effective workspace visualization to a remote mentor. The visualization is stable, complete, up to date, continuous, distortion free, and rendered from the mentee’s typical viewpoint, as needed to best inform the mentor of the current state of the workspace. In one study, the stabilized visualization had significant advantages over unstabilized visualization, in the context of three number matching tasks. In a second study, stabilization showed good results, in the context of surgical telementoring, specifically for cricothyroidotomy training in austere settings.

Datasets are an essential component for training effective machine learning models. In particular, surgical robotic datasets have been key to many advances in semi-autonomous surgeries, skill assessment, and training. Simulated surgical environments can enhance the data collection process by making it faster, simpler and cheaper than real systems. In addition, combining data from multiple robotic domains can provide rich and diverse training data for transfer learning algorithms. In this paper, we present the DESK (DExterous Surgical SKills) dataset. It comprises a set of surgical robotic skills collected during a surgical training task using three robotic platforms: the Taurus II robot, Taurus II simulated robot, and the YuMi robot. This dataset was used to test the idea of transferring knowledge across different domains (e.g. from Taurus to YuMi robot) for a surgical gesture classification task with seven gestures/surgemes. We explored two different scenarios: 1) No transfer and 2) Domain transfer (simulated Taurus to real Taurus and YuMi robots). We conducted extensive experiments with three supervised learning models and provided baselines in each of these scenarios. Results show that using simulation data during training enhances the performance on the real robots, where limited real data is available. In particular, we obtained an accuracy of 55% on the real Taurus data using a model that is trained only on the simulator data, but that accuracy improved to 82% when the ratio of real to simulated data was increased to 0.18 in the training set.

In the future, deployable, teleoperated surgical robots can save the lives of critically injured patients in battlefield environments. These robotic systems will need to have autonomous capabilities to take over during communication delays and unexpected environmental conditions during critical phases of the procedure. Understanding and predicting the next surgical actions (referred as “surgemes”) is essential for autonomous surgery. Most approaches for surgeme recognition cannot cope with the high variability associated with austere environments and thereby cannot “transfer” well to field robotics. We propose a methodology that uses compact image representations with kinematic features for surgeme recognition in the DESK dataset. This dataset offers samples for surgical procedures over different robotic platforms with a high variability in the setup. We performed surgeme classification in two setups: 1) No transfer, 2) Transfer from a simulated scenario to two real deployable robots. Then, the results were compared with recognition accuracies using only kinematic data with the same experimental setup. The results show that our approach improves the recognition performance over kinematic data across different domains. The proposed approach produced a transfer accuracy gain up to 20% between the simulated and the real robot, and up to 31% between the simulated robot and a different robot. A transfer accuracy gain was observed for all cases, even those already above 90%.

This poster presents the design, implementation, and evaluation of a method for robust high-level stabilization of mentees first-person video in augmented reality (AR) telementoring. This video is captured by the front-facing built-in camera of an AR headset and stabilized by rendering from a stationary view a planar proxy of the workspace projectively texture mapped with the video feed. The result is stable, complete, up to date, continuous, distortion free, and rendered from the mentee's default viewpoint. The stabilization method was evaluated in two user studies, in the context of number matching and for cricothyroidotomy training, respectively. Both showed a significant advantage of our method compared with unstabilized visualization.

User's intentions may be expressed through spontaneous gesturing, which have been seen only a few times or never before. Recognizing such gestures involves one shot gesture learning. While most research has focused on the recognition of the gestures themselves, recently new approaches were proposed to deal with gesture perception and production as part of the recognition problem. The framework presented in this work focuses on learning the process that leads to gesture generation, rather than treating the gestures as the outcomes of a stochastic process only. This is achieved by leveraging kinematic and cognitive aspects of human interaction. These factors enable the artificial production of realistic gesture samples originated from a single observation, which in turn are used as training sets for state-of-the-art classifiers. Classification performance is evaluated in terms of recognition accuracy and coherency; the latter being a novel metric that determines the level of agreement between humans and machines. Specifically, the referred machines are robots which perform artificially generated examples. Coherency in recognition was determined at 93.8%, corresponding to a recognition accuracy of 89.2% for the classifiers and 92.5% for human participants. A proof of concept was performed towards the expansion of the proposed one shot learning approach to adaptive learning, and the results are presented and the implications discussed.