Katherine J. Kuchenbecker directs the Haptic Intelligence Department at the Max Planck Institute for Intelligent Systems in Stuttgart, Germany. She was previously an Associate Professor of Mechanical Engineering and Applied Mechanics at the University of Pennsylvania, where she held the Class of 1940 Bicentennial Endowed Term Chair and a secondary appointment in Computer and Information Science. Kuchenbecker earned her Ph.D. in Mechanical Engineering at Stanford University in 2006 and did a postdoctoral fellowship at the Johns Hopkins University. Her research centers on haptic interfaces, which enable a user to touch virtual and distant objects as though they were real and within reach, as well as haptic sensing systems, which allow robots to physically interact with objects and people.
Kuchenbecker delivered a TEDYouth talk on haptics in 2012, and she has received several honors including a 2009 NSF CAREER Award, the 2012 IEEE Robotics and Automation Society Academic Early Career Award, and a 2014 Penn Lindback Award for Distinguished Teaching. Her team has won various best paper and best demonstration awards, and she frequently gives keynote talks at conferences. She was co-chair of the IEEE Technical Committee on Haptics from 2014 to 2017, and she co-chaired the IEEE Haptics Symposium in 2016 and 2018.
In Proceedings of the IEEE World Haptics Conference (WHC), pages: 107-112, June 2017 (inproceedings)
Over time, surgical trainees learn to compensate for the lack of haptic feedback in commercial robotic minimally invasive surgical systems. Incorporating touch cues into robotic surgery training could potentially shorten this learning process if the benefits of haptic feedback were sustained after it is removed. In this paper, we develop a wrist-squeezing haptic feedback system and evaluate whether it holds the potential to train novice da Vinci users to reduce the force they exert on a bimanual inanimate training task. Subjects were randomly divided into two groups according to a multiple baseline experimental design. Each of the ten participants moved a ring along a curved wire nine times while the haptic feedback was conditionally withheld, provided, and withheld again. The realtime tactile feedback of applied force magnitude significantly reduced the integral of the force produced by the da Vinci tools on the task materials, and this result remained even when the haptic feedback was removed. Overall, our findings suggest that wrist-squeezing force feedback can play an essential role in helping novice trainees learn to minimize the force they exert with a surgical robot.
In Proceedings of the IEEE World Haptics Conference (WHC), pages: 599-604, June 2017, Finalist for best poster paper (inproceedings)
Despite rapid advancements in the field of fingertip haptics, rendering tactile cues with six degrees of freedom (6 DOF) remains an elusive challenge. In this paper, we investigate the potential of displaying fingertip haptic sensations with a 6-DOF parallel continuum manipulator (PCM) that mounts to the user's index finger and moves a contact platform around the fingertip. Compared to traditional mechanisms composed of rigid links and discrete joints, PCMs have the potential to be strong, dexterous, and compact, but they are also more complicated to design. We define the design space of 6-DOF parallel continuum manipulators and outline a process for refining such a device for fingertip haptic applications. Following extensive simulation, we obtain 12 designs that meet our specifications, construct a manually actuated prototype of one such design, and evaluate the simulation's ability to accurately predict the prototype's motion. Finally, we demonstrate the range of deliverable fingertip tactile cues, including a normal force into the finger and shear forces tangent to the finger at three extreme points on the boundary of the fingertip.
In Proceedings of the IEEE World Haptics Conference (WHC), pages: 394-399, June 2017 (inproceedings)
Clever electromechanical design is required to make the force feedback delivered by a kinesthetic haptic interface both strong and safe. This paper explores a onedimensional haptic force display that combines a DC motor and a magnetic particle brake on the same shaft. Rather than a rigid linkage, a spooled cable connects the user to the actuators to enable a large workspace, reduce the moving mass, and eliminate the sticky residual force from the brake. This design combines the high torque/power ratio of the brake and the active output capabilities of the motor to provide a wider range of forces than can be achieved with either actuator alone. A prototype of this device was built, its performance was characterized, and it was used to simulate constant force sources and virtual springs and dampers. Compared to the conventional design of using only a motor, the hybrid device can output higher unidirectional forces at the expense of free space feeling less free.
In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), pages: 439-445, May 2017 (inproceedings)
The Portable Robotic Optical/Tactile ObservatioN PACKage (PROTONPACK, or Proton for short) is a new handheld visuo-haptic sensing system that records surface interactions. We previously demonstrated system calibration and a classification task using external motion tracking. This paper details improvements in surface classification performance and removal of the dependence on external motion tracking, necessary before embarking on our goal of gathering a vast surface interaction dataset. Two experiments were performed to refine data collection parameters. After adjusting the placement and filtering of the Proton's high-bandwidth accelerometers, we recorded interactions between two differently-sized steel tooling ball end-effectors (diameter 6.35 and 9.525 mm) and five surfaces. Using features based on normal force, tangential force, end-effector speed, and contact vibration, we trained multi-class SVMs to classify the surfaces using 50 ms chunks of data from each end-effector. Classification accuracies of 84.5% and 91.5% respectively were achieved on unseen test data, an improvement over prior results. In parallel, we pursued on-board motion tracking, using the Proton's camera and fiducial markers. Motion tracks from the external and onboard trackers agree within 2 mm and 0.01 rad RMS, and the accuracy decreases only slightly to 87.7% when using onboard tracking for the 9.525 mm end-effector. These experiments indicate that the Proton 2 is ready for portable data collection.
Short paper presented at the Workshop on C4 Surgical Robots at the IEEE International Conference on Robotics and Automation (ICRA), May 2017 (misc)
Teleoperated surgical robots such as the Intuitive da Vinci Surgical System facilitate minimally invasive surgeries, which decrease risk to patients. However, these systems can be difficult to learn, and existing training curricula on surgical simulators do not offer students the realistic experience of a full operation. This paper presents an augmented-reality video training platform for the da Vinci that will allow trainees to rehearse any surgery recorded by an expert. While the trainee operates a da Vinci in free space, they see their own instruments overlaid on the expert video. Tools are identified in the source videos via color segmentation and kernelized correlation filter tracking, and their depth is calculated from the da Vinci’s stereoscopic video feed. The user tries to follow the expert’s movements, and if any of their tools venture too far away, the system provides instantaneous visual feedback and pauses to allow the user to correct their motion. The trainee can also rewind the expert video by bringing either da Vinci tool very close to the camera. This combined and augmented video provides the user with an immersive and interactive training experience.
Hands-on demonstration presented at ACM/IEEE International Conference on Human-Robot Interaction (HRI), March 2017 (misc)
Robots that work alongside humans might be more effective if they could forge a strong social bond with their human partners. Hand-clapping games and other forms of rhythmic social-physical interaction may foster human-robot teamwork, but the design of such interactions has scarcely been explored. At the HRI 2017 conference, we will showcase several such interactions taken from our recent work with the Rethink Robotics Baxter Research Robot, including tempo-matching, Simon says, and Pat-a-cake-like games. We believe conference attendees will be both entertained and intrigued by this novel demonstration of social-physical HRI.
Surgical Endoscopy, Proceedings of the 2017 Annual Meeting of the Society of American Gastrointestinal and Endoscopic Surgeons (SAGES), 31(Supplement 1):S28, Springer, March 2017 (article)
Introduction: Minimally invasive surgery has revolutionized surgical practice, but challenges remain. Trainees must acquire complex technical skills while minimizing patient risk, and surgeons must maintain their skills for rare procedures. These challenges are magnified in pediatric surgery due to the smaller spaces, finer tissue, and relative dearth of both inanimate and virtual simulators. To build technical expertise, trainees need opportunities for deliberate practice with specific performance feedback, which is typically provided via tedious human grading. This study aimed to validate a novel motion-tracking system and machine learning algorithm for automatically evaluating trainee performance on a pediatric laparoscopic suturing task using a 1–5 OSATS Overall Skill rating.
Methods: Subjects (n=14) ranging from medical students to fellows per- formed one or two trials of an intracorporeal suturing task in a custom pediatric laparoscopy training box (Fig. 1) after watching a video of ideal performance by an expert. The position and orientation of the tools and endoscope were recorded over time using Ascension trakSTAR magnetic motion-tracking sensors, and both instrument grasp angles were recorded over time using flex sensors on the handles. The 27 trials were video-recorded and scored on the OSATS scale by a senior fellow; ratings ranged from 1 to 4. The raw motion data from each trial was processed to calculate over 200 preliminary motion parameters. Regularized least-squares regression (LASSO) was used to identify the most predictive parameters for inclusion in a regression tree. Model performance was evaluated by leave-one-subject-out cross validation, wherein the automatic scores given to each subject’s trials (by a model trained on all other data) are compared to the corresponding human rater scores.
Results: The best-performing LASSO algorithm identified 14 predictive parameters for inclusion in the regression tree, including completion time, linear path length, angular path length, angular acceleration, grasp velocity, and grasp acceleration. The final model’s raw output showed a strong positive correlation of 0.87 with the reviewer-generated scores, and rounding the output to the nearest integer yielded a leave-one-subject-out cross-validation accuracy of 77.8%. Results are summarized in the confusion matrix (Table 1).
Conclusions: Our novel motion-tracking system and regression model automatically gave previously unseen trials overall skill scores that closely match scores from an expert human rater. With additional data and further development, this system may enable creation of a motion-based training platform for pediatric laparoscopic surgery and could yield insights into the fundamental components of surgical skill.
Workshop paper (5 pages) presented at the AAAI Spring Symposium on Interactive Multi-Sensory Object Perception for Embodied Agents, March 2017 (misc)
The Proton Pack is a portable visuo-haptic surface interaction recording device that will be used to collect a vast multimodal dataset, intended for robots to use as part of an approach to understanding the world around them. In order to collect a useful dataset, we want to pick a suitable interaction duration for each surface, noting the tradeoff between data collection resources and completeness of data. One interesting approach frames the data collection process as an online learning problem, building an incremental surface model and using that model to decide when there is enough data. Here we examine how to do such online surface modeling and when to stop collecting data, using kinetic friction as a first domain in which to apply online modeling.
In Social Robotics: 8th International Conference, ICSR 2016, Kansas City, MO, USA, November 1-3, 2016 Proceedings, 9979, pages: 317-327, Lecture Notes in Artificial Intelligence, Springer International Publishing, November 2016, Oral presentation given by Fitter (inproceedings)
In Social Robotics: 8th International Conference, ICSR 2016, Kansas City, MO, USA, November 1-3, 2016 Proceedings, 9979, pages: 340-350, Lecture Notes in Artificial Intelligence, Springer International Publishing, November 2016, Oral presentation given by Fitter (inproceedings)
In Social Robotics: 8th International Conference, ICSR 2016, Kansas City, MO, USA, November 1-3, 2016 Proceedings, 9979, pages: 296-305, Lecture Notes in Artificial Intelligence, Springer International Publishing, November 2016, Oral presentation given by Fitter (inproceedings)
Our goal is to understand the principles of Perception, Action and Learning in autonomous systems that successfully interact with complex environments and to use this understanding to design future systems