Juan Wachs

Deepak Raina, SH Chandrashekhara, Richard Voyles et al.

Ultrasound is a commonly used medical imaging modality that requires expert sonographers to manually maneuver the ultrasound probe based on the acquired image. Autonomous Robotic Ultrasound (A-RUS) is an appealing alternative to this manual procedure in order to reduce sonographers' workload. The key challenge to A-RUS is optimizing the ultrasound image quality for the region of interest across different patients. This requires knowledge of anatomy, recognition of error sources and precise probe position, orientation and pressure. Sample efficiency is important while optimizing these parameters associated with the robotized probe controller. Bayesian Optimization (BO), a sample-efficient optimization framework, has recently been applied to optimize the 2D motion of the probe. Nevertheless, further improvements are needed to improve the sample efficiency for high-dimensional control of the probe. We aim to overcome this problem by using a neural network to learn a low-dimensional kernel in BO, termed as Deep Kernel (DK). The neural network of DK is trained using probe and image data acquired during the procedure. The two image quality estimators are proposed that use a deep convolution neural network and provide real-time feedback to the BO. We validated our framework using these two feedback functions on three urinary bladder phantoms. We obtained over 50% increase in sample efficiency for 6D control of the robotized probe. Furthermore, our results indicate that this performance enhancement in BO is independent of the specific training dataset, demonstrating inter-patient adaptability.

Deepak Raina, Abhishek Mathur, Richard M. Voyles et al.

The one of the significant challenges faced by autonomous robotic ultrasound systems is acquiring high-quality images across different patients. The proper orientation of the robotized probe plays a crucial role in governing the quality of ultrasound images. To address this challenge, we propose a sample-efficient method to automatically adjust the orientation of the ultrasound probe normal to the point of contact on the scanning surface, thereby improving the acoustic coupling of the probe and resulting image quality. Our method utilizes Bayesian Optimization (BO) based search on the scanning surface to efficiently search for the normalized probe orientation. We formulate a novel objective function for BO that leverages the contact force measurements and underlying mechanics to identify the normal. We further incorporate a regularization scheme in BO to handle the noisy objective function. The performance of the proposed strategy has been assessed through experiments on urinary bladder phantoms. These phantoms included planar, tilted, and rough surfaces, and were examined using both linear and convex probes with varying search space limits. Further, simulation-based studies have been carried out using 3D human mesh models. The results demonstrate that the mean ($\pm$SD) absolute angular error averaged over all phantoms and 3D models is $\boldsymbol{2.4\pm0.7^\circ}$ and $\boldsymbol{2.1\pm1.3^\circ}$, respectively.

Maxwell J. Jacobson, Daniela Chanci Arrubla, Maria Romeo Tricas et al.

Approximately 1.25 million people in the United States are treated each year for burn injuries. Precise burn injury classification is an important aspect of the medical AI field. In this work, we propose an explainable human-in-the-loop framework for improving burn ultrasound classification models. Our framework leverages an explanation system based on the LIME classification explainer to corroborate and integrate a burn expert's knowledge -- suggesting new features and ensuring the validity of the model. Using this framework, we discover that B-mode ultrasound classifiers can be enhanced by supplying textural features. More specifically, we confirm that texture features based on the Gray Level Co-occurance Matrix (GLCM) of ultrasound frames can increase the accuracy of transfer learned burn depth classifiers. We test our hypothesis on real data from porcine subjects. We show improvements in the accuracy of burn depth classification -- from ~88% to ~94% -- once modified according to our framework.

Yutong Zhang, Shaiv Y. Mehra, Bradley S. Duerstock et al.

While powered wheelchairs reduce physical fatigue as opposed to manual wheelchairs for individuals with mobility impairment, they demand high cognitive workload due to information processing, decision making and motor coordination. Current autonomous systems lack social awareness in navigation and transparency in decision-making, leading to decreased perceived safety and trust from the user and others in context. This work proposes Socially Aware Autonomous Transparent Transportation (SAATT) Navigation framework for wheelchairs as a potential solution. By implementing a Large Language Model (LLM) informed of user intent and capable of predicting other peoples' intent as a decision-maker for its local controller, it is able to detect and navigate social situations, such as passing pedestrians or a pair conversing. Furthermore, the LLM textually communicates its reasoning at each waypoint for transparency. In this experiment, it is compared against a standard global planner, a representative competing social navigation model, and an Ablation study in three simulated environments varied by social levels in eight metrics categorized under Safety, Social Compliance, Efficiency, and Comfort. Overall, SAATT Nav outperforms in most social situations and equivalently or only slightly worse in the remaining metrics, demonstrating the potential of a socially aware and transparent autonomous navigation system to assist wheelchair users.

Deepak Raina, Mythra V. Balakuntala, Byung Wook Kim et al.

Ultrasound is widely employed for clinical intervention and diagnosis, due to its advantages of offering non-invasive, radiation-free, and real-time imaging. However, the accessibility of this dexterous procedure is limited due to the substantial training and expertise required of operators. The robotic ultrasound (RUS) offers a viable solution to address this limitation; nonetheless, achieving human-level proficiency remains challenging. Learning from demonstrations (LfD) methods have been explored in RUS, which learns the policy prior from a dataset of offline demonstrations to encode the mental model of the expert sonographer. However, active engagement of experts, i.e. Coaching, during the training of RUS has not been explored thus far. Coaching is known for enhancing efficiency and performance in human training. This paper proposes a coaching framework for RUS to amplify its performance. The framework combines DRL (self-supervised practice) with sparse expert's feedback through coaching. The DRL employs an off-policy Soft Actor-Critic (SAC) network, with a reward based on image quality rating. The coaching by experts is modeled as a Partially Observable Markov Decision Process (POMDP), which updates the policy parameters based on the correction by the expert. The validation study on phantoms showed that coaching increases the learning rate by $25\%$ and the number of high-quality image acquisition by $74.5\%$.

Chenxi Xiao, Shujia Xu, Wenzhuo Wu et al.

Robotic exploration under uncertain environments is challenging when optical information is not available. In this paper, we propose an autonomous solution of exploring an unknown task space based on tactile sensing alone. We first designed a whisker sensor based on MEMS barometer devices. This sensor can acquire contact information by interacting with the environment non-intrusively. This sensor is accompanied by a planning technique to generate exploration trajectories by using mere tactile perception. This technique relies on a hybrid policy for tactile exploration, which includes a proactive informative path planner for object searching, and a reactive Hopf oscillator for contour tracing. Results indicate that the hybrid exploration policy can increase the efficiency of object discovery. Last, scene understanding was facilitated by segmenting objects and classification. A classifier was developed to recognize the object categories based on the geometric features collected by the whisker sensor. Such an approach demonstrates the whisker sensor, together with the tactile intelligence, can provide sufficiently discriminative features to distinguish objects.

Mythra V. Balakuntala, Upinder Kaur, Xin Ma et al.

Everyday contact-rich tasks, such as peeling, cleaning, and writing, demand multimodal perception for effective and precise task execution. However, these present a novel challenge to robots as they lack the ability to combine these multimodal stimuli for performing contact-rich tasks. Learning-based methods have attempted to model multi-modal contact-rich tasks, but they often require extensive training examples and task-specific reward functions which limits their practicality and scope. Hence, we propose a generalizable model-free learning-from-demonstration framework for robots to learn contact-rich skills without explicit reward engineering. We present a novel multi-modal sensor data representation which improves the learning performance for contact-rich skills. We performed training and experiments using the real-life Sawyer robot for three everyday contact-rich skills -- cleaning, writing, and peeling. Notably, the framework achieves a success rate of 100\% for the peeling and writing skill, and 80\% for the cleaning skill. Hence, this skill learning framework can be extended for learning other physical manipulation skills.

Anirudh Tunga, Sai Vidyaranya Nuthalapati, Juan Wachs

Sign language recognition (SLR) plays a crucial role in bridging the communication gap between the hearing and vocally impaired community and the rest of the society. Word-level sign language recognition (WSLR) is the first important step towards understanding and interpreting sign language. However, recognizing signs from videos is a challenging task as the meaning of a word depends on a combination of subtle body motions, hand configurations, and other movements. Recent pose-based architectures for WSLR either model both the spatial and temporal dependencies among the poses in different frames simultaneously or only model the temporal information without fully utilizing the spatial information. We tackle the problem of WSLR using a novel pose-based approach, which captures spatial and temporal information separately and performs late fusion. Our proposed architecture explicitly captures the spatial interactions in the video using a Graph Convolutional Network (GCN). The temporal dependencies between the frames are captured using Bidirectional Encoder Representations from Transformers (BERT). Experimental results on WLASL, a standard word-level sign language recognition dataset show that our model significantly outperforms the state-of-the-art on pose-based methods by achieving an improvement in the prediction accuracy by up to 5%.

Glebys T. Gonzalez, Upinder Kaur, Masudur Rahma et al.

Short response time is critical for future military medical operations in austere settings or remote areas. Such effective patient care at the point of injury can greatly benefit from the integration of semi-autonomous robotic systems. To achieve autonomy, robots would require massive libraries of maneuvers. While this is possible in controlled settings, obtaining surgical data in austere settings can be difficult. Hence, in this paper, we present the Dexterous Surgical Skill (DESK) database for knowledge transfer between robots. The peg transfer task was selected as it is one of 6 main tasks of laparoscopic training. Also, we provide a ML framework to evaluate novel transfer learning methodologies on this database. The collected DESK dataset comprises a set of surgical robotic skills using the four robotic platforms: Taurus II, simulated Taurus II, YuMi, and the da Vinci Research Kit. Then, we explored two different learning scenarios: no-transfer and domain-transfer. In the no-transfer scenario, the training and testing data were obtained from the same domain; whereas in the domain-transfer scenario, the training data is a blend of simulated and real robot data that is tested on a real robot. Using simulation data enhances the performance of the real robot where limited or no real data is available. The transfer model showed an accuracy of 81% for the YuMi robot when the ratio of real-to-simulated data was 22%-78%. For Taurus II and da Vinci robots, the model showed an accuracy of 97.5% and 93% respectively, training only with simulation data. Results indicate that simulation can be used to augment training data to enhance the performance of models in real scenarios. This shows the potential for future use of surgical data from the operating room in deployable surgical robots in remote areas.

Glebys Gonzalez, Juan Wachs

Achieving human-like motion in robots has been a fundamental goal in many areas of robotics research. Inverse kinematic (IK) solvers have been explored as a solution to provide kinematic structures with anthropomorphic movements. In particular, numeric solvers based on geometry, such as FABRIK, have shown potential for producing human-like motion at a low computational cost. Nevertheless, these methods have shown limitations when solving for robot kinematic constraints. This work proposes a framework inspired by FABRIK for human pose imitation in real-time. The goal is to mitigate the problems of the original algorithm while retaining the resulting humanlike fluidity and low cost. We first propose a human constraint model for pose imitation. Then, we present a pose imitation algorithm (PIC), and it's soft version (PICs) that can successfully imitate human poses using the proposed constraint system. PIC was tested on two collaborative robots (Baxter and YuMi). Fifty human demonstrations were collected for a bi-manual assembly and an incision task. Then, two performance metrics were obtained for both robots: pose accuracy with respect to the human and the percentage of environment occlusion/obstruction. The performance of PIC and PICs was compared against the numerical solver baseline (FABRIK). The proposed algorithms achieve a higher pose accuracy than FABRIK for both tasks (25%-FABRIK, 53%-PICs, 58%-PICs). In addition, PIC and it's soft version achieve a lower percentage of occlusion during incision (10%-FABRIK, 4%-PICs, 9%-PICs). These results indicate that the PIC method can reproduce human poses and achieve key desired effects of human imitation.

Edgar Rojas-Muñoz, Kyle Couperus, Juan Wachs

Telementoring surgeons as they perform surgery can be essential in the treatment of patients when in situ expertise is not available. Nonetheless, expert mentors are often unavailable to provide trainees with real-time medical guidance. When mentors are unavailable, a fallback autonomous mechanism should provide medical practitioners with the required guidance. However, AI/autonomous mentoring in medicine has been limited by the availability of generalizable prediction models, and surgical procedures datasets to train those models with. This work presents the initial steps towards the development of an intelligent artificial system for autonomous medical mentoring. Specifically, we present the first Database for AI Surgical Instruction (DAISI). DAISI leverages on images and instructions to provide step-by-step demonstrations of how to perform procedures from various medical disciplines. The dataset was acquired from real surgical procedures and data from academic textbooks. We used DAISI to train an encoder-decoder neural network capable of predicting medical instructions given a current view of the surgery. Afterwards, the instructions predicted by the network were evaluated using cumulative BLEU scores and input from expert physicians. According to the BLEU scores, the predicted and ground truth instructions were as high as 67% similar. Additionally, expert physicians subjectively assessed the algorithm using Likert scale, and considered that the predicted descriptions were related to the images. This work provides a baseline for AI algorithms to assist in autonomous medical mentoring.

Chenxi Xiao, Juan Wachs

Three dimensional (3D) object recognition is becoming a key desired capability for many computer vision systems such as autonomous vehicles, service robots and surveillance drones to operate more effectively in unstructured environments. These real-time systems require effective classification methods that are robust to various sampling resolutions, noisy measurements, and unconstrained pose configurations. Previous research has shown that points' sparsity, rotation and positional inherent variance can lead to a significant drop in the performance of point cloud based classification techniques. However, neither of them is sufficiently robust to multifactorial variance and significant sparsity. In this regard, we propose a novel approach for 3D classification that can simultaneously achieve invariance towards rotation, positional shift, scaling, and is robust to point sparsity. To this end, we introduce a new feature that utilizes graph structure of point clouds, which can be learned end-to-end with our proposed neural network to acquire a robust latent representation of the 3D object. We show that such latent representations can significantly improve the performance of object classification and retrieval tasks when points are sparse. Further, we show that our approach outperforms PointNet and 3DmFV by 35.0% and 28.1% respectively in ModelNet 40 classification tasks using sparse point clouds of only 16 points under arbitrary SO(3) rotation.

Naveen Madapana, Glebys Gonzalez, Juan Wachs

Participatory design is a popular design technique that involves the end users in the early stages of the design process to obtain user-friendly gestural interfaces. Guessability studies followed by agreement analyses are often used to elicit and comprehend the preferences (or gestures/proposals) of the participants. Previous approaches to assess agreement, grouped the gestures into equivalence classes and ignored the integral properties that are shared between them. In this work, we represent the gestures using binary description vectors to allow them to be partially similar. In this context, we introduce a new metric referred to as soft agreement rate (SAR) to quantify the level of consensus between the participants. In addition, we performed computational experiments to study the behavior of our partial agreement formula and mathematically show that existing agreement metrics are a special case of our approach. Our methodology was evaluated through a gesture elicitation study conducted with a group of neurosurgeons. Nevertheless, our formulation can be applied to any other user-elicitation study. Results show that the level of agreement obtained by SAR metric is 2.64 times higher than the existing metrics. In addition to the mostly agreed gesture, SAR formulation also provides the mostly agreed descriptors which can potentially help the designers to come up with a final gesture set.

Mythra V. Balakuntala, Vishnunandan L. N. Venkatesh, Jyothsna Padmakumar Bindu et al.

Humans generally teach their fellow collaborators to perform tasks through a small number of demonstrations. The learnt task is corrected or extended to meet specific task goals by means of coaching. Adopting a similar framework for teaching robots through demonstrations and coaching makes teaching tasks highly intuitive. Unlike traditional Learning from Demonstration (LfD) approaches which require multiple demonstrations, we present a one-shot learning from demonstration approach to learn tasks. The learnt task is corrected and generalized using two layers of evaluation/modification. First, the robot self-evaluates its performance and corrects the performance to be closer to the demonstrated task. Then, coaching is used as a means to extend the policy learnt to be adaptable to varying task goals. Both the self-evaluation and coaching are implemented using reinforcement learning (RL) methods. Coaching is achieved through human feedback on desired goal and action modification to generalize to specified task goals. The proposed approach is evaluated with a scooping task, by presenting a single demonstration. The self-evaluation framework aims to reduce the resistance to scooping in the media. To reduce the search space for RL, we bootstrap the search using least resistance path obtained using resistive force theory. Coaching is used to generalize the learnt task policy to transfer the desired quantity of material. Thus, the proposed method provides a framework for learning tasks from one demonstration and generalizing it using human feedback through coaching.

Naveen Madapana, Md Masudur Rahman, Natalia Sanchez-Tamayo et al.

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 Skill) 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. We explored three different scenarios: 1) No transfer, 2) Transfer from simulated Taurus to real Taurus and 3) Transfer from Simulated Taurus to the YuMi robot. 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 robot 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. Furthermore, we achieved an accuracy improvement of 34% when 3% of the real data is added into the training process.

Tian Zhou, Juan P. Wachs

This paper introduces the Turn-Taking Spiking Neural Network (TTSNet), which is a cognitive model to perform early turn-taking prediction about human or agent's intentions. The TTSNet framework relies on implicit and explicit multimodal communication cues (physical, neurological and physiological) to be able to predict when the turn-taking event will occur in a robust and unambiguous fashion. To test the theories proposed, the TTSNet framework was implemented on an assistant robotic nurse, which predicts surgeon's turn-taking intentions and delivers surgical instruments accordingly. Experiments were conducted to evaluate TTSNet's performance in early turn-taking prediction. It was found to reach a F1 score of 0.683 given 10% of completed action, and a F1 score of 0.852 at 50% and 0.894 at 100% of the completed action. This performance outperformed multiple state-of-the-art algorithms, and surpassed human performance when limited partial observation is given (< 40%). Such early turn-taking prediction capability would allow robots to perform collaborative actions proactively, in order to facilitate collaboration and increase team efficiency.

Tian Zhou, Juan P. Wachs

Turn-taking is essential to the structure of human teamwork. Humans are typically aware of team members' intention to keep or relinquish their turn before a turn switch, where the responsibility of working on a shared task is shifted. Future co-robots are also expected to provide such competence. To that end, this paper proposes the Cognitive Turn-taking Model (CTTM), which leverages cognitive models (i.e., Spiking Neural Network) to achieve early turn-taking prediction. The CTTM framework can process multimodal human communication cues (both implicit and explicit) and predict human turn-taking intentions in an early stage. The proposed framework is tested on a simulated surgical procedure, where a robotic scrub nurse predicts the surgeon's turn-taking intention. It was found that the proposed CTTM framework outperforms the state-of-the-art turn-taking prediction algorithms by a large margin. It also outperforms humans when presented with partial observations of communication cues (i.e., less than 40% of full actions). This early prediction capability enables robots to initiate turn-taking actions at an early stage, which facilitates collaboration and increases overall efficiency.

Tian Zhou, Juan P. Wachs

To enable a natural and fluent human robot collaboration flow, it is critical for a robot to comprehend their human peers' on-going actions, predict their behaviors in the near future, and plan its actions correspondingly. Specifically, the capability of making early predictions is important, so that the robot can foresee the precise timing of a turn-taking event and start motion planning and execution early enough to smooth the turn-taking transition. Such proactive behavior would reduce human's waiting time, increase efficiency and enhance naturalness in collaborative task. To that end, this paper presents the design and implementation of an early turn-taking prediction algorithm, catered for physical human robot collaboration scenarios. Specifically, a Robotic Scrub Nurse (RSN) system which can comprehend surgeon's multimodal communication cues and perform turn-taking prediction is presented. The developed algorithm was tested on a collected data set of simulated surgical procedures in a surgeon-nurse tandem. The proposed turn-taking prediction algorithm is found to be significantly superior to its algorithmic counterparts, and is more accurate than human baseline when little partial input is given (less than 30% of full action). After observing more information, the algorithm can achieve comparable performances as humans with a F1 score of 0.90.

Tian Zhou, Maria E. Cabrera, Juan P. Wachs

This study tries to explain the connection between communication modalities and levels of supervision in teleoperation during a dexterous task, like surgery. This concept is applied to two surgical related tasks: incision and peg transfer. It was found that as the complexity of the task escalates, the combination linking human supervision with a more expressive modality shows better performance than other combinations of modalities and control. More specifically, in the peg transfer task, the combination of speech modality and action level supervision achieves shorter task completion time (77.1 +- 3.4 s) with fewer mistakes (0.20 +- 0.17 pegs dropped).

Maria Cabrera, Richard Voyles, Juan Wachs

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 itself, recently new approaches were proposed to deal with gesture perception and production as part of the same problem. The framework presented in this work focuses on learning the process that leads to gesture generation, rather than mining the gesture's associated features. This is achieved using kinematic, cognitive and biomechanic characteristics of human interaction. These factors enable the artificial production of realistic gesture samples originated from a single observation. The generated samples are then used as training sets for different state-of-the-art classifiers. Performance is obtained first, by observing the machines' gesture recognition percentages. Then, performance is computed by the human recognition from gestures performed by robots. Based on these two scenarios, a composite new metric of coherency is proposed relating to the amount of agreement between these two conditions. Experimental results provide an average recognition performance of 89.2% for the trained classifiers and 92.5% for the participants. Coherency in recognition was determined at 93.6%. While this new metric is not directly comparable to raw accuracy or other pure performance-based standard metrics, it provides a quantifier for validating how realistic the machine generated samples are and how accurate the resulting mimicry is.

Maria Cabrera, Keisha Novak, Daniel Foti et al.

Previous work in the area of gesture production, has made the assumption that machines can replicate "human-like" gestures by connecting a bounded set of salient points in the motion trajectory. Those inflection points were hypothesized to also display cognitive saliency. The purpose of this paper is to validate that claim using electroencephalography (EEG). That is, this paper attempts to find neural signatures of gestures (also referred as placeholders) in human cognition, which facilitate the understanding, learning and repetition of gestures. Further, it is discussed whether there is a direct mapping between the placeholders and kinematic salient points in the gesture trajectories. These are expressed as relationships between inflection points in the gestures' trajectories with oscillatory mu rhythms (8-12 Hz) in the EEG. This is achieved by correlating fluctuations in mu power during gesture observation with salient motion points found for each gesture. Peaks in the EEG signal at central electrodes (motor cortex) and occipital electrodes (visual cortex) were used to isolate the salient events within each gesture. We found that a linear model predicting mu peaks from motion inflections fits the data well. Increases in EEG power were detected 380 and 500ms after inflection points at occipital and central electrodes, respectively. These results suggest that coordinated activity in visual and motor cortices is sensitive to motion trajectories during gesture observation, and it is consistent with the proposal that inflection points operate as placeholders in gesture recognition.