Core Concepts
AO-Grasp introduces a method to generate stable and actionable 6 DoF grasps on articulated objects directly from partial point clouds.
Abstract
AO-Grasp proposes a novel grasp proposal method for robots to interact with articulated objects, such as cabinets and appliances. It consists of the AO-Grasp Model and Dataset, achieving higher success rates than baselines in simulation and real-world scenarios. The model predicts grasp points using an Actionable Grasp Point Predictor without requiring part detection or hand-designed heuristics. Training on the new dataset enables the model to generate stable and actionable grasps on diverse objects with varied geometries and articulation axes. AO-Grasp demonstrates zero-shot sim-to-real transfer capabilities, outperforming existing methods in both simulated and real-world environments.
Stats
AO-Grasp achieves a 45.0% grasp success rate in simulation.
In real-world scenes, AO-Grasp produces successful grasps on 67.5% of instances.
Quotes
"AO-Grasp is the first method for generating 6 DoF grasps on articulated objects directly from partial point clouds."
"AO-Grasp achieves higher success rates than baselines in both simulation and real-world scenarios."