October 2017 – September 2021
The goal of Alex’s research is the creation of a robust intention recognition system for a mobile robot operating in an agricultural setting. The agricultural domain poses additional challenges like changing lighting, large distances, and higher variety in user behavior compared to common Human-Robot Interaction settings. We aim to overcome these challenges by using a combination of measures. Firstly, to overcome individual sensor limitations, we utilize a number of different sensors. We integrate their data into a combined model of the human pose before using the integrated pose model as a basis for action labeling. Secondly, to balance the expected ambiguity of detected actions, we consider these actions in the context of situational information (time, location, close objects) as well as past experience with the individual in question. The combined approach should increase action detection robustness.