Using collaborative robots in a workshop environment can alleviate strains on workers and increase productivity, but improvements are still required for workers to want to work with them.
Pick-and-place tasks are one of the most widespread applications for robots, but they are still not as efficient as human workers when it comes to selecting grasps for different objects.
In this thesis, a method is presented that leverages the human worker's intuition of optimal grasps to teach the robot which grasp to use for different objects.
A Convolutional Neural Network has been developed to classify between 6 different grasps with an accuracy of 87.22%, while objects are recognised using ORB features with an accuracy of 56.61%.
The method has been tested with 9 test subjects using a UR3 robot, and their feedback of the system regarding comfort and trust in the implementation resulted in a score of 5.89 on a 7-point Likert scale, showing that there is an interest in collaborating with a robot in this manner.