Horizontal Multi-Surface Random Sample Consensus for Robust Customizable Shelf Perception

Abstract

Many robot tasks across domains from ware- houses to homes, such as object fetching, reconfiguration, and stowing, involve interactions with shelves. In this paper, we present a variant of the Random Sample Consensus (RANSAC) algorithm, with a sampling strategy that exploits the regular structure of shelves to robustly detect them. We examined the efficacy and usability of our algorithm by comparing it to a state-of-the-art surface segmentation algorithm implemented in PCL both with general parameters and with shelf-specific customized parameters. The results indicate that our algorithm has better precision and much higher recall in both settings.

BibTeX Entry

@inproceedings{peng2019irosw,
  title = {Horizontal Multi-Surface Random Sample Consensus for Robust Customizable Shelf Perception},
  author = {Peng, Yu-Tang and Huang, Justin and Cakmak, Maya},
  year = {2019},
  month = nov,
  booktitle = {IROS 2019 Workshop on Different Approaches, the Same Goal: Autonomous Object Manipulation},
  type = {workshop}
}