Robotic object rearrangement combines the skills of picking and placing objects. When object models are unavailable, typical collision-checking models may be unable to predict collisions in partial point clouds with occlusions, making generation of collision-free grasping or placement trajectories challenging. We propose a learned collision model that accepts scene and query object point clouds and predicts collisions for 6DOF object poses within the scene. We train the model on a synthetic set of 1 million scene/object point cloud pairs and 2 billion collision queries. We leverage the learned collision model as part of a model predictive path integral (MPPI) policy in a tabletop rearrangement task and show that the policy can plan collision-free grasps and placements for objects unseen in training in both simulated and physical cluttered scenes with a Franka Panda robot. The learned model outperforms both traditional pipelines and learned ablations by 9.8% in accuracy on a dataset of simulated collision queries and is 75x faster than the best-performing baseline.

Three examples of multiple barrier scenarios where the task is to pick an object by itself on one side of the barrier and place it (possibly among other objects) on the other side of the barrier. The larger barriers mean that successful trajectories must further deviate from the straight-line trajectory than even the single barrier case. In the third video, we show an example of a case where the centroid of the object point cloud lies beyond the bounds of the collision checker due to the height of the object, which results in incorrect collision predictions.

If you use this paper or code in your work, please consider citing: