Parameter Identification for a Non-modular Elastic Joint Robot Arm for Observer-based Collision Detection

Safe physical human-robot interaction gains importance since bringing humans and robots spatially working together provides a high benefit for industry. Here robots can aid humans e.g. as "third hand" or by performing monotonous tasks. To realize this, a certain level of safety has to be ensured. A lightweight design and inherent passive compliance, as present at the BioRobArm, helps to reduce the injury risk, which is caused by a robot. Beside this, various performed collision tests from present research showed that a reliable collision detection can reduce the collision force in many cases, or at least dissipate dangerous situations for an involved human. This work implements a model-based disturbance observer for collision detection. Since an accurate model had to be available to ensure a reliable collision detection, a general approach to produce the dynamics model for robots with elastic joints is introduced. To use the model as observer, its parameters have to be identified. For this purpose a method is proposed, that combines two approaches, which separately treat the actuator and load side model to create a linear equation system. These equation systems are then solved using excitation trajectories, which are optimized according to an appropriate observability measure. The model identification process is verified by simulations and experiments. Finally the implemented model-based collision detection is successfully tested during a collision test with an appropriate reaction. The tests have shown that the proposed methods can be used for model identification and collision detection, but the produced model has to be refined to better represent the real behavior. Also the benefit of the collision detection has to be evaluated in further tests with the real robot.