Palpation is a commonly used diagnostic process in which a clinician examines the mechanical properties of human tissue via tactile exploration. Here we are interested in a system that can guide and train clinicians to do palpation more efficiently. We proposed a partially observable markov decision process (POMDP) framework for modelling the robotic guidance in the palpation task. We tested the proposed algorithm with a few use cases and showed that it can correctly identify situations where the operator needs guidance.

In collaborative tasks, people usually have different preferences, and thus develop different strategies for collaboration. For example, in a two person task, they may become leader and follower, or they may equally share the work. In this project, I tried to classify human preference and "type" using force and trajectory data from an experiment in my lab. I extracted features from the raw data, and performed PCA and clustering. The analysis revealed two major collaboration patterns, which matched the observation from the original experiment.