Benchmarking the Effects of Object Pose Estimation and Reconstruction on Robotic Grasping Success
Varun Burde,, Pavel Burget, Torsten Sattler
AI summary
Problem
Standard geometric metrics for 3D reconstruction and 6D pose estimation fail to capture how perception errors compound and impact downstream robotic manipulation tasks like grasping.
Approach
The authors introduce a large-scale physics-based simulation benchmark that evaluates how errors from 6D pose estimation and 3D mesh reconstruction propagate to affect robotic grasp success across multiple grippers and objects.
Key results
- Large-scale physics-based benchmark linking perception errors to grasp success
- Reconstruction artifacts significantly reduce viable grasp candidates but negligibly affect success when pose is accurate
- Grasping success is dominated by 3D spatial pose error, with translation error strongly predicting success for symmetric objects
- Gripper choice heavily influences success rates and failure modes, with no single gripper optimal for all objects
Why it matters
Provides robotics researchers and perception developers with a functional evaluation paradigm to understand how geometric inaccuracies impact real-world manipulation performance.
Abstract
3D reconstruction serves as the foundational layer for numerous robotic perception tasks, including 6D object pose estimation and grasp pose generation. Modern 3D reconstruc- tion methods for objects can produce visually and geometrically impressive meshes from multi-view images, yet standard geo- metric evaluations do not reflect how reconstruction quality influences downstream tasks such as robotic manipulation performance. This paper addresses this gap by introducing a large-scale, physics-based benchmark that evaluates 6D pose estimators and 3D mesh models based on their functional efficacy in grasping. We analyze the impact of model fidelity by generating grasps on various reconstructed 3D meshes and executing them on a high-fidelity reference model, simulating how grasp poses generated with an imperfect reconstruction affect physical interaction. This assesses the combined impact of pose error, grasp robustness, and geometric inaccuracies from 3D reconstruction. Our results show that reconstruction artifacts significantly decrease the number of grasp pose candi- dates but have a negligible effect on grasping performance given an accurately estimated pose. Our results also reveal that the relationship between grasp success and pose error is dominated by spatial error, and even a simple translation error provides insight into the success of the grasping pose of symmetric objects. This work provides insight into how perception systems relate to object manipulation using robots.