Collision Detection and Distinction from Expected Contacts in Human-Robot Cooperation (Kouris, A. - 2016)


As human-robot cooperation is becoming more and more widespread in the last decades, safety issues are raised considering the fact that a human and a robot are sharing the same workspace. The ability of a robot to detect collisions within its environment, distinguish them from intended contacts that are present during cooperative tasks and identify their characteristics (such as the collision point and the direction), is necessary to be embedded in all robots participating in such cooperative tasks, in order to reduce the risk of an accident.

In this work, we review methods that have been proposed in the bibliography to address this issue, some of which are implemented and tested. Following that, a new method is proposed, taking into consideration the desirable characteristics of the robot’s dynamic behaviour during the cooperation and covers an open issue of robotics research. This new method is based on processing the measurement of the external force applied on the robot during cooperation in the frequency domain. We also propose techniques for identifying the characteristics of the detected collision, based on our method.

Designing appropriate reactions for the robot in order to minimise the impact of a collision, is of major interest in order to avoid or reduce human injury. The main target of the design is to achieve human safety during human-robot cooperation and coexistence in the same workspace. Secondarily it is important to ensure that the proposed techniques do not downgrade the robot’s performance by limiting its accuracy, speed and real-time interaction with its environment. By utilizing the information provided by the proposed method, we design reaction strategies that target in meeting the above criteria, with common property that all proposed strategies are based on the same control scheme that the robot was using during normal cooperation. This allows the human and the robot to jointly ensure human safety.

The proposed methods are generalised to be appropriate for multiple directions and to identify collisions taking place on the entire robot’s body and are being tested on a KUKA LWR 4+, a seven DOF robot manipulator. The tests show that the proposed method achieves improved distinction between intended contacts and unexpected collisions of the robot with its environment, compared to examined methods from the literature. The collision detection time is significantly reduced, concluding into faster reaction that leads in minimising the applied force from the robot on the contact point, during the collision.



Thesis PDF