Collaborative and Cooperative Robotics Applications using Visual Perception

The objective of this Thesis is to develop novel integrated strategies for collaborative and cooperative robotic applications. Commonly, industrial robots operate in structured environments and in work-cell separated from human operators. Nowadays, collaborative robots have the capacity of sharing the workspace and collaborate with humans or other robots to perform complex tasks. These robots often operate in an unstructured environment, whereby they need sensors and algorithms to get information about environment changes. Advanced vision and control techniques have been analyzed to evaluate their performance and their applicability to industrial tasks. Then, some selected techniques have been applied for the first time to an industrial context. A Peg-in-Hole task has been chosen as first case study, since it has been extensively studied but still remains challenging: it requires accuracy both in the determination of the hole poses and in the robot positioning. Two solutions have been developed and tested. Experimental results have been discussed to highlight the advantages and disadvantages of each technique. Grasping partially known objects in unstructured environments is one of the most challenging issues in robotics. It is a complex task and requires to address multiple subproblems, in order to be accomplished, including object localization and grasp pose detection. Also for this class of issues some vision techniques have been analyzed. One of these has been adapted to be used in industrial scenarios. Moreover, as a second case study, a robot-to-robot object handover task in a partially structured environment and in the absence of explicit communication between the robots has been developed and validated. Finally, the two case studies have been integrated in two real industrial setups to demonstrate the applicability of the strategies to solving industrial problems.