Human-aware Collaborative Manipulation with Reaching Motion Prediction

• Main Author: Fioravanço, Lucas Monteiro
• Publication Date: 2021
• Format: Master thesis
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10362/125779
• topic: Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

This dissertations presents a possible approach to improve human-robot interaction in an industrial collaborative situation, where the human operator and a collaborative industrial robot work within a shared work-space. The approach presented in this dissertation focuses on a situation where part of the assembly process needs to be carried out by a human operator, whose assembly station is located on a work-bench, and a robot is used to pick and place products in specific locations on the operator’s work station. Because those locations can be accessed both by the robot or the human operator at any time, collisions can occur and should be avoided in order to make the process more natural for the human operator as well as to avoid the emergency stop of the collaborative robot which has to be restarted and thus decreases productivity. In order to prevent those collisions the proposed system defines key-areas in each of the locations as well as other relevant positions for the collaborative task. The system uses a Kinect Sensor and a neural network to track the user’s hand over time and Gaussian Mixture Models to make predictions regarding the possible destination key-area given the observed trajectory until that moment. If a collision is predicted the robot pauses the task being executed at the moment in order to prevent it and, once the conflict has been resolved, resumes operation.