# Conditional Visual Servoing for Multi-Step Tasks

**Authors:** Sergio Izquierdo, Max Argus, Thomas Brox  
**arXiv:** [2205.08441](https://arxiv.org/abs/2205.08441)  
**Date:** 2022-05-17  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2205.08441v1/

## Key points

- Paper introduces conditional visual servoing for multi-step tasks
- Modular approach called conditional servoing allows for flexibility in tackling multi-step problems
- Next demonstration chosen based on observation of the robot
- Approach compared on shape-sorting task in simulation using different selection functions
- Reprojection error yields best overall results
- Selection function implemented on real robot and demonstrated to be effective
- Strategy extends visual servoing to handle scenarios with multiple demonstration sequences
- Promising results in both simulation and real-world experiments

## Summary

The paper titled "Conditional Visual Servoing for Multi-Step Tasks" by Sergio Izquierdo, Max Argus, and Thomas Brox introduces a novel approach to extend visual servoing to scenarios with multiple demonstration sequences. The authors propose a modular approach called conditional servoing that allows for more flexibility in tackling multi-step problems. This is achieved by choosing the next demonstration based on the observation of the robot. The proposed approach is compared on a shape-sorting task in simulation using different selection functions, with the reprojection error yielding the best overall results. To validate their findings, the authors implement this selection function on a real robot and demonstrate its efficacy. Overall, this paper presents an innovative strategy for extending visual servoing to handle scenarios with multiple demonstration sequences and offers promising results in both simulation and real-world experiments. For more information and videos of their experiments, readers are directed to visit the project page: https://lmb.informatik.uni-freiburg.de/projects/conditional_servoing/.