Human-Robot Collaboration for Kinesthetic Teaching

Recent industrial interest in producing smaller volumes of products in shorter time frames, in contrast to mass production in previous decades, motivated the introduction of human–robot collaboration (HRC) in industrial settings, as an attempt to increase flexibility in manufacturing applications by incorporating human intelligence and dexterity to these processes. This thesis presents methods for improving the involvement of human operators in industrial settings where robots are present, with a particular focus on kinesthetic teaching, i.e., manually guiding the robot to define or correct its motion, since it can facilitate non-expert robot programming.
To increase flexibility in the manufacturing industry implies a loss of a fixed structure of the industrial environment, which increases the uncertainties in the shared workspace between humans and robots. Two methods have been proposed in this thesis to mitigate such uncertainty. First, null-space motion was used to increase the accuracy of kinesthetic teaching by reducing the joint static friction, or stiction, without altering the execution of the robotic task. This was possible since robots used in HRC, i.e., collaborative robots, are often designed with additional degrees of freedom (DOFs) for a greater dexterity. Second, to perform effective corrections of the motion of the robot through kinesthetic teaching in partially-unknown industrial environments, a fast identification of the source of robot–environment contact is necessary. Fast contact detection and classification methods in literature were evaluated, extended, and modified to use them in kinesthetic teaching applications for an assembly task. For this, collaborative robots that are made compliant with respect to their external forces/torques (as an active safety mechanism) were used, and only embedded sensors of the robot were considered.
Moreover, safety is a major concern when robotic motion occurs in an inherently uncertain scenario, especially if humans are present. Therefore, an online variation of the compliant behavior of the robot during its manual guidance by a human operator was proposed to avoid undesired parts of the workspace of the robot. The proposed method used safety control barrier functions (SCBFs) that considered the rigid-body dynamics of the robot, and the method’s stability was guaranteed using a passivity-based energy-storage formulation that includes a strict Lyapunov function.
All presented methods were tested experimentally on a real collaborative robot.