Hao-Tien Lewis Chiang

Qiping Zhang, Nathan Tsoi, Mofeed Nagib et al.

Understanding how humans evaluate robot behavior during human-robot interactions is crucial for developing socially aware robots that behave according to human expectations. While the traditional approach to capturing these evaluations is to conduct a user study, recent work has proposed utilizing machine learning instead. However, existing data-driven methods require large amounts of labeled data, which limits their use in practice. To address this gap, we propose leveraging the few-shot learning capabilities of Large Language Models (LLMs) to improve how well a robot can predict a user's perception of its performance, and study this idea experimentally in social navigation tasks. To this end, we extend the SEAN TOGETHER dataset with additional real-world human-robot navigation episodes and participant feedback. Using this augmented dataset, we evaluate the ability of several LLMs to predict human perceptions of robot performance from a small number of in-context examples, based on observed spatio-temporal cues of the robot and surrounding human motion. Our results demonstrate that LLMs can match or exceed the performance of traditional supervised learning models while requiring an order of magnitude fewer labeled instances. We further show that prediction performance can improve with more in-context examples, confirming the scalability of our approach. Additionally, we investigate what kind of sensor-based information an LLM relies on to make these inferences by conducting an ablation study on the input features considered for performance prediction. Finally, we explore the novel application of personalized examples for in-context learning, i.e., drawn from the same user being evaluated, finding that they further enhance prediction accuracy. This work paves the path to improving robot behavior in a scalable manner through user-centered feedback.

Robert Grupp, Hsin-Hong Chiang, Yoshito Otake et al.

Several methods to perform extrapolation of unknown anatomy were evaluated. The primary application is to enhance surgical procedures that may use partial medical images or medical images of incomplete anatomy. Le Fort-based, face-jaw-teeth transplant is one such procedure. From CT data of 36 skulls and 21 mandibles separate Statistical Shape Models of the anatomical surfaces were created. Using the Statistical Shape Models, incomplete surfaces were projected to obtain complete surface estimates. The surface estimates exhibit non-zero error in regions where the true surface is known; it is desirable to keep the true surface and seamlessly merge the estimated unknown surface. Existing extrapolation techniques produce non-smooth transitions from the true surface to the estimated surface, resulting in additional error and a less aesthetically pleasing result. The three extrapolation techniques evaluated were: copying and pasting of the surface estimate (non-smooth baseline), a feathering between the patient surface and surface estimate, and an estimate generated via a Thin Plate Spline trained from displacements between the surface estimate and corresponding vertices of the known patient surface. Feathering and Thin Plate Spline approaches both yielded smooth transitions. However, feathering corrupted known vertex values. Leave-one-out analyses were conducted, with 5% to 50% of known anatomy removed from the left-out patient and estimated via the proposed approaches. The Thin Plate Spline approach yielded smaller errors than the other two approaches, with an average vertex error improvement of 1.46 mm and 1.38 mm for the skull and mandible respectively, over the baseline approach.