Cobot Learning from mid-level representations to low-level robot execution skills (TEAMSTER)

Forskningsprojekt, 2024 – 2028

Robotics and automation have shown tremendous success in various sectors, such as manufacturing, food service, mining, etc., however, in other important areas, such as assembly or maintenance, where collaborative tasks between humans and robots are required, robotics solutions have shown limited results. Collaborative tasks require that two or more agents work together as a team to solve a common goal, for example, assembling production kits. Collaborative Robots (Cobots) are intended to interact directly with humans in shared and dynamic spaces. Meaning that humans and robots can work together in close proximity to achieve a common goal. What is limiting the deployment of Cobots to complex collaborative scenarios? Is it the lack of a unified representation of the environment (states)? Can robots use their previous experience (e.g. modeled as a graph) to learn and predict the next action to assist humans? The purpose of this project is to develop an interpretable learning method to enable Cobots to predict and select the optimal actions from a learned control library (skills) to facilitate collaborations, mirroring human's ability to make quick and informed decisions in dynamic collaborative settings. Thus, Cobots will learn a high-level reasoning model to resemble human cognitive modeling, and a low-level control model to execute optimal skills. The synthesis of high-level reasoning with adequate low-level controls is very challenging and an unexplored territory that will open up new opportunities in different research areas within the HRC field.