# Compact Dual-Quaternion-Based Visual Servoing for Perpendicular Alignment with Surface Normal Constraints

## TL;DR

This paper introduces a new visual servoing method using dual quaternions to improve robotic button-pressing accuracy by ensuring perpendicular alignment.

## Contribution

A novel dual-quaternion-based visual servoing method is proposed for accurate and efficient robotic pressing with surface normal constraints.

## Key findings

- The method ensures perpendicular alignment between the end-effector and target surface during pressing.
- Experimental results show improved positioning accuracy and computational efficiency compared to conventional methods.

## Abstract

The ability to reliably press physical buttons is a common requirement in robotics. Conventional vision-based methods often suffer from positional errors during execution if the end-effector’s approach is not perpendicular to the target surface. This paper proposes a novel dual-quaternion-based visual servoing method that enables robots to reach desired poses and enhances accuracy in robotic button-pressing. Our method acquires target pose information (position, depth and surface normal direction) from the RGB-D camera and converts it into dual quaternion representation to construct the visual servoing control system. The image Jacobian matrix for the dual quaternion pose is then computed. The dual-quaternion-based visual servoing ensures that the pressing direction and the optical axis of the coaxially mounted camera remain perpendicular throughout the pressing motion, thereby eliminating misalignment between the actual contact point and the visually identified target. By representing spatial displacements in SE(3) with dual quaternions, our method enables more compact, concise, and efficient pose representation and computation throughout the visual servoing process. Experimental results demonstrate that, compared to conventional methods, our technique achieves more efficient visual servoing control, significantly improving both positioning accuracy and computational efficiency.