Reactive pre-collision strategies

From a control point of view this monograph dealt to a large extent with physical collisions, their detection and following reaction up to now. Apart from such physical analysis and control, immanent injury can be diminished if the robot is able to reduce its impact speed or change its moving direction prior to the collision. Locally, the robot would circumvent the human or obstacle and avoid the impact completely. Therefore, it is of major importance to provide flexible motion generation methods, which take into account the possibly complex environment structure and at the same time can react quickly to changing conditions. Motion generation methods can be divided into path planning algorithms and reactive motion generation. On the one hand (probabilistic) complete, highly sophisticated offline path planning methods are used, which provide complete collision free paths for potentially complex scenarios [4] with multi-DoF open or closed chain kinematics. On the other hand, reactive motion generators, which usually show a more responsive behavior, are simpler and have short execution cycles. Usually, these methods associate virtual forces to obstacles that act on virtual dynamics assigned to the robot. Both classes mostly treat the entire motion generation problem from a purely geometric/kinematic point of view. However, with the recent advances in pHRI it becomes more important to be able to plan complex motions for task achievement and cope with the proximity of dynamic obstacles under the absolute premise of safety to the human at the same time. However, under these constraints both existing approaches have significant drawbacks. Complex motion planners cannot match the real-time requirements of the low-level control cycle due to their computational complexity. Reactive methods on the other hand do usually not provide completeness and are (some more, others less) prone to get stuck in local minima. Most importantly however, both approaches do not incorporate physical forces into their according behavior. Therefore, they are not able to treat forces not as a failure but as an additional sensory input that provides valuable information. This dilemma necessitates to treat motion planning, collision avoidance, and collision detection/reaction in a unified approach.