Physics-Based Grasp Planning Through Clutter

We propose a planning method for grasping in cluttered environments, where the robot can make simultaneous contact with multiple objects, manipulating them in a deliberate and controlled fashion. This enables the robot to reach for and grasp the target while simultaneously contacting and moving aside obstacles in order to clear a desired path. We use a physicsbased analysis of pushing to compute the motion of each object in the scene in response to a set of possible robot motions. In order to make the problem computationally tractable, we enable multiple simultaneous robot-object interactions, which we pre-compute and cache, but avoid object-object interactions. Tests on large sets of simulated scenes show that our planner produces more successful grasps in more complex scenes than versions that avoid any interaction with surrounding clutter. Validation on a real robot shows that our grasp evaluation method accurately predicts the outcome of a grasp, and that our approach, in conjunction with state-of-the-art object recognition tools, is applicable in reallife scenes that are highly cluttered and constrained.