Differential Amplifier-Inspired AmpAttention for Multi-View Robotic Manipulation

2026-07-02

AI Summary

This work addresses the challenges of multi-view robotic manipulation, where image redundancy, occlusions, and viewpoint dependency often lead to attention drift. To mitigate these issues, the authors propose AmpAttention, a novel mechanism that, for the first time, incorporates the differential amplification principle from analog circuits into attention computation. By leveraging task-guided intra- and inter-view attention, AmpAttention suppresses noise and enhances perceptual signal-to-noise ratio. Built upon this mechanism, the RVAF model enables efficient and robust multi-view fusion. An enhanced variant, RVAF++, built on the SAM2 image encoder, achieves state-of-the-art average success rates across 18 RLBench tasks (249 variants) while reducing training time by 33.3%. Notably, it attains a 91% success rate on the “peg insertion” task and successfully executes a high-precision real-world dart-throwing task targeting the bullseye.

Abstract

Multi-view robotic manipulation methods with the attention mechanism have recently achieved significant progress in both training efficiency and task performance. However, the inherent redundancy, occlusion, and viewpoint dependency in robotic view images often lead to severe attention drift. To address this challenge, we propose AmpAttention, a novel attention mechanism inspired by differential amplifiers in analog circuits. It aims to suppress attention noise and capture high signal-to-noise ratio signals for more reliable perception. Based on this, we introduce the RVAF model, which integrates task-guided intra-view and inter-view AmpAttention. Compared to previous state-of-the-art methods, RVAF achieves the optimal average success rate across 18 RLBench tasks (249 variations) while reducing training time by 33.3\%. RVAF also demonstrates strong potential in real-world high-precision tasks, exemplified by its ability to pick up a dart and accurately insert it into the red bullseye. Furthermore, we extend RVAF to RVAF++ by incorporating the SAM2 image encoder. RVAF++ achieves substantial gains on high-precision tasks, achieving a 91\% success rate on the `insert peg' task. More qualitative results are provided at the anonymous project website https://anonymous.4open.science/w/RVAF-Anonymization.

Problem

• attention drift
• multi-view robotic manipulation
• occlusion
• viewpoint dependency
• redundancy

Innovation

• AmpAttention
• differential amplifier-inspired attention
• multi-view robotic manipulation
• attention drift suppression
• RVAF