aiqbench analysis
This research addresses critical training instabilities in 3D policy learning by introducing a novel architecture that couples transformer-based encoders with diffusion decoders to improve scalability and robustness. By diagnosing specific failure modes like the lack of 3D data augmentation and adverse Batch Normalization effects, the work establishes a stronger foundation for cross-embodiment transfer in robotic manipulation tasks.
R3D: Revisiting 3D Policy Learning
arXiv:2604.15281v2 Announce Type: replace-cross Abstract: 3D policy learning promises superior generalization and cross-embodiment transfer, but progress has been hindered by training instabilities and severe overfitting, precluding the adoption of powerful 3D perception models. In this work, we systematically diagnose these failures, identifying the omission of 3D data augmentation and the adverse effects of Batch Normalization as primary causes. We propose a new architecture coupling a scalable transformer-based 3D encoder with a diffusion decoder, engineered specifically for stability at scale and designed to leverage large-scale pre-training. Our approach significantly outperforms state-of-the-art 3D baselines on challenging manipulation benchmarks, establishing a new and robust foundation for scalable 3D imitation learning. Project Page: https://r3d-policy.github.io/
Abstract:3D policy learning promises superior generalization and cross-embodiment transfer, but progress has been hindered by training instabilities and severe overfitting, precluding the adoption of powerful 3D perception models. In this work, we systematically diagnose these failures, identifying the omission of 3D data augmentation and the adverse effects of Batch Normalization as primary causes. We propose a new architecture coupling a scalable transformer-based 3D encoder with a diffusion decoder, engineered specifically for stability at scale and designed to leverage large-scale pre-training. Our approach significantly outperforms state-of-the-art 3D baselines on challenging manipulation benchmarks, establishing a new and robust foundation for scalable 3D imitation learning. Project Page: this https URL
Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
arXiv:2604.15281 [cs.CV]
(or arXiv:2604.15281v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2604.15281 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 204 words
Related coverage
The Machines that Make the Machines
NVIDIA · Oct 7, 2026
Revisiting the "Push-T" Robot Manipulation Task with Agentic Robotics
arXiv · Aug 20, 2026
Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning
arXiv · Oct 9, 2026
Beyond Single-Model Optimization: Preserving Plasticity in Continual Reinforcement Learning
arXiv · Jun 10, 2026