An investigation on the effects of mass, velocity, and surface geometry on human soft-tissue injury, with applications to knowledge-based real-time robot control.

Making robots able to safetly interact with human is one of the most important goals in robotics resarch. The developments achieved on the last years in mechanical design and control opened up the possibility for novel tasks that require active cooperation between human and robot. For such applications safety, dependability and adaptability to human behavior become the most important requirements to be guaranteed in order to allow the cooperation. In order to understand what safe behavior means, the understanding of possible injury caused by a robot is foundamental. The objective of this thesis is therefore to examine soft-tissue behavior during collisions and investigate the role of robot mass, velocity, and surface geometry on injury severity. In the end, using these tests results, a safe velocity-control strategy is proposed that ensures human safety even in the worse case of unexpected constrained collision, i.e. the impact velocities remain always below potentially dangerous values. In the first part of the thesis an overview of the current state of the art in pHRI is given. The most important hardware design and control strategies developed up to now are briefly described. An elencation of the current regulations in human-robot interaction is given and a summary of crash-tests in robotics is provided. Finally, an overview of the physical properties of soft-tissue from medical and forensical literature is presented. In the second part of the thesis, the setup and testing procedure of the drop testing on swine subjects are described in detail. Then, after an investigation on the sensor data results and the medical analysis of the specimens, a procedure for the evaluation of "safety curves" for identifying "safe" robot behavior in terms of speed, is developed. The architecture of an "injury database" that holds the respective parameterization of these curves (in order to make them available in real time for robot speed control) is defined. Finally, a control scheme for limiting the end-effector velocity based on the safety curves evaluated from tests, is designed, implemented, and experimentally verified. This scheme takes into account the reflected behavior at the respective point of interest on the end-effector (in terms of inertia and velocity) by evaluating the possible injury that could be caused in real-time