VolumeDP: Modeling Volumetric Representation for Manipulation Policy Learning

TL;DR

VolumeDP introduces a 3D-aware policy architecture for robotic manipulation that explicitly reasons in volumetric space, significantly improving success rates and robustness over existing 2D-to-3D mapping methods.

Contribution

The paper proposes VolumeDP, a novel volumetric representation-based policy architecture that enhances spatial reasoning and robustness in robotic manipulation tasks.

Findings

Achieves 88.8% success rate on LIBERO benchmark, outperforming baselines by 14.8%.

Demonstrates superior performance on ManiSkill and LIBERO-Plus benchmarks.

Shows improved real-world generalization to new spatial layouts and viewpoints.

Abstract

Imitation learning is a prominent paradigm for robotic manipulation. However, existing visual imitation methods map 2D image observations directly to 3D action outputs, imposing a 2D-3D mismatch that hinders spatial reasoning and degrades robustness. We present VolumeDP, a policy architecture that restores spatial alignment by explicitly reasoning in 3D. VolumeDP first lifts image features into a Volumetric Representation via cross-attention. It then selects task-relevant voxels with a learnable module and converts them into a compact set of spatial tokens, markedly reducing computation while preserving action-critical geometry. Finally, a multi-token decoder conditions on the entire token set to predict actions, thereby avoiding lossy aggregation that collapses multiple spatial tokens into a single descriptor. VolumeDP achieves a state-of-the-art average success rate of 88.8% on the LIBERO simulation benchmark, outperforming the strongest baseline by a substantial 14.8% improvement. It also delivers large performance gains over prior methods on the ManiSkill and LIBERO-Plus benchmarks. Real-world experiments further demonstrate higher success rates and robust generalization to novel spatial layouts, camera viewpoints, and environment backgrounds. Code will be released.