IPAB Workshop - 23/06/22

Speaker: Emile Mackute

Title: Shape Estimation of Concentric Tube Robots Using Single Point Position Measurement

Abstract: Accurate shape estimation of concentric tube robots (CTRs) using mathematical models remains a challenge, reinforcing the need to develop techniques for accurate and real-time shape sensing of CTRs. In this talk I will present a sensor fusion algorithm that predicts and tracks the robot's shape by combining a mathematical model of the CTR with an electromagnetic sensor measurement of the Cartesian coordinates of the robot's tip. We experimentally validated our method in static and dynamic scenarios with and without external loading, demonstrating that the sensor fusion algorithm improves the error of model-based shape prediction by an average of 44.3%. Furthermore, we showed that this approach can be used in real-time.

Speaker: Ahmed Raza

Title: EatSense: Recognition and Quality Assessment of Fine-Grained Actions for Eating Activities

Abstract: The elderly population is increasing at a rapid rate and the need for effectively supporting independent living has become crucial. Wearable sensors can be helpful, but these are intrusive as they require adherence by the elderly. Thus, a semi-anonymous (no image records) vision-based non-intrusive monitoring system might potentially be the answer. As everyone has to eat, we introduce a first investigation into how eating behavior might be used as an indicator of behavior change. We present a method for detailed analysis of eating behavior and a novel vision-based dataset collected while people eat. For an in-depth understanding of eating and changes in behavior over time, we (i) label the common sub-actions, (ii) simulate deterioration in the movement by adding weight to the wrists. The proposed approach can recognize actions with 74.4% on a frame-by-frame basis with the high-level features extracted from 3D joint detection and tracking. We can also distinguish between decayed and normal (with and without weight) motion with accuracy >95% when training on mixed data (data from all subjects).