Event
DefFlow Improves Deformable Object Manipulation Success Rate with Joint 3D Point Flow
1 reports · 1 sources
Gaurav Singh and other researchers propose the DefFlow framework, which jointly predicts goal-conditioned 3D point flow for the robot and the deformable object from demonstrations, covering decisions such as grasping, releasing, and regrasping as well as continuous actions toward the goal. Across various initial and goal configurations of ropes, granular media, and cloth, DefFlow significantly outperforms goal-conditioned action prediction and robot-flow-only baselines on success rate and goal-distance metrics. The framework has been deployed on a real robot using real point cloud observations and unseen goal configurations.
Report timeline
- DefFlow: A Goal-Conditioned Framework for Deformable Object Manipulation Based on 3D Point Flows
DefFlow is a goal-conditioned manipulation framework for deformable objects. By jointly predicting the goal-conditioned 3D point flows of the robot and the deformable object from demonstrations, it captures manipulation decisions such as grasping, releasing, and re-grasping, as well as the continuous fine motions required to reach the goal. Across rope, granular media, and cloth settings with diverse initial and target configurations, DefFlow clearly outperforms goal-conditioned action prediction and robot-flow-only baselines on success rate and goal-distance metrics, and it was deployed on a real robot using real point cloud observations and unseen target configurations.
arXiv cs.RO ·