Stable Object Reorientation using Contact Plane Registration
This work addresses a key challenge in robotics for object manipulation, offering incremental improvements in accuracy and transferability.
The paper tackles the problem of predicting stable orientations for rigid objects from noisy, partial pointclouds, achieving substantial performance improvements over state-of-the-art systems in simulated stacking tasks and demonstrating strong sim2real zero-shot transfer on real-world reorientation tasks.
We present a system for accurately predicting stable orientations for diverse rigid objects. We propose to overcome the critical issue of modelling multimodality in the space of rotations by using a conditional generative model to accurately classify contact surfaces. Our system is capable of operating from noisy and partially-observed pointcloud observations captured by real world depth cameras. Our method substantially outperforms the current state-of-the-art systems on a simulated stacking task requiring highly accurate rotations, and demonstrates strong sim2real zero-shot transfer results across a variety of unseen objects on a real world reorientation task. Project website: \url{https://richardrl.github.io/stable-reorientation/}