Action Map Policy: Learning 3D Closed-loop Manipulation via Pixel Classification

2026-07-12

AI Summary

This work addresses the challenge of efficiently and accurately modeling high-dimensional, long-horizon, and multimodal robotic action spaces, which are difficult for conventional methods to handle. The authors propose reframing 3D closed-loop manipulation policy learning as a pixel-wise classification problem in image space, where 3D actions are projected onto the camera image plane and represented as discrete action categories corresponding to pixel locations. This formulation enables end-to-end prediction while maintaining multimodality and millimeter-level spatial precision, all within a constrained action vocabulary. Notably, it allows for the generation of complete action sequences in a single forward pass. Experimental results demonstrate that the proposed approach significantly outperforms existing baselines across multiple manipulation tasks, achieving consistent improvements in success rate, inference speed, and spatial reasoning capability.

Abstract

The action space poses a major challenge in robot learning, since it is often high-dimensional, can span long time horizons, and frequently admits multi-modal optimal solutions. A good choice of action representation and loss function can help to address these concerns, but there are often trade offs. We propose Action Map Policy (AMP), which casts 3D closed-loop manipulation policy learning as a classification problem in image space. While classification has been an effective formulation in generative language models, applying it to robot action learning is difficult because naively discretizing high-dimensional continuous actions explodes the token vocabulary. Our key idea is to project 3D actions onto the camera image planes and treat each pixel location as a discrete class, thus controlling dimensionality while retaining multi-modality. This method supports millimeter-level precision for high-dimensional actions without requiring a prohibitively large vocabulary, while preserving fine-grained pixel-wise visual signals. Furthermore, it can predict the entire action chunk in a single forward pass, avoiding complex noise scheduling and iterative denoising while achieving substantially faster inference than diffusion policies. Experiments on various manipulation tasks show that AMP outperforms strong baselines, achieving higher success rates, faster inference, and enhanced spatial reasoning.

Problem

• action space
• high-dimensional actions
• multi-modal solutions
• robot learning
• 3D manipulation

Innovation

• Action Map Policy
• pixel classification
• 3D manipulation
• closed-loop control
• action discretization