Grasp selection for simple manipulation tasks

Manipulation tasks are sequential in nature. Depending on the specific task constraints, more than one solution usually exist. Even for simple objects, several candidate grasps can be considered, each of which can potentially be used to plan the required arm motions. Grasp selection approaches that take into account the constraints at each task step are critical, since they allow to both identify grasps that are more likely to produce feasible arm motions and analogously, to discard grasps that might not be executable due to future task constraints. In this abstract, we present a manipulation metric for grasp selection based on a combined arm-and-grasp measure evaluated at each task step. We analyze 3 simple tasks (pick-up, pick-and-place and pouring tasks) and show the advantages of using our metric: (1) Shorter end-effector displacements and (2) Higher planning success rates. We present quantitative results in simulation and validate our approach’s practicality with experimental results in our physical robot platform.