A robot task can be represented as a set of trajectories conformed by a sequence of poses. In this way it is possible to teach a mobile robot to accomplish a manipulation task, and also to reproduce it. Nevertheless robot navigation may normally introduce inaccuracies in localization due to natural events as wheel-slides, causing a mismatch between the end-effector and the objects or tools the robot is supposed to interact with. We propose an algorithm for adapting manipulation trajectories for different locations. The adaptation is achieved by optimizing in position, orientation and energy consumption. The approach is built over the basis of Evolution Strategies, and only uses forward kinematics permitting to avoid all the inconveniences that inverse kinematics imply, as well as convergence problems in singular kinematic configurations. Manipulation paths generated with this algorithm can achieve optimal performance, sometimes even improving original path smoothness. Experimental results are presented to verify the algorithm.
展开▼
