What is it about?
The effect of approximation has been well studied in stand-alone systems. However, the problem of approximation in a collaborative system has not been studied, to the best of our knowledge. The basic goal of this article is to study the applicability of approximation in collaborative SLAM (simultaneous localization and mapping). Our experiments suggest that it is not trivial to combine multiple stand-alone approximate results to achieve a collaborative approximation, i.e., the resultant error can not be bounded without special effort. Thus, we present a model of the problem and empirically show that such a model can be used to explain the error variance in a collaborative system.
Featured Image
Photo by Gavin Allanwood on Unsplash
Why is it important?
Modern generalized computing systems are over-provisioned for accuracy, so they spend more than the necessary time and energy to achieve a łcorrectž output. In recent years, inexact or approximate computing has emerged as an alternative technique to address resource constraints in embedded systems. The underlying philosophy is to sacrifice a little accuracy from computation and gain in terms of time or energy or both. There are a variety of applications that are inherently tolerant of approximation, e.g., computer vision, media processing, machine learning, etc. A comprehensive study on approximate computing, in both the fields of hardware and software, can be found in the literature.
Perspectives
We study the behavior of approximate computing on the MR-SLAMalgorithm in the context of the multi-robot system. We present a model for approximation in the collaborative system. Our experiments uphold the model and show that the approximation error increases as the number of collaborators increases in the system. Also, there are some specific execution paths in which the error magnifies. To avoid such paths, in order to reduce error in the overall outcome, the agents in the system need to coordinate and hence communicate more. Since communication cost for mobile robots is significantly less than the motion cost, this would not impact negatively on the overall efficiency of the system. Our future work includes modeling such interaction and optimizing the trade-off between communication and error in the outcome of the collaborative system.
Hrishav Bakul Barua
TCS Research
Read the Original
This page is a summary of: A study of approximation in a collaborative multi-agent system, January 2020, ACM (Association for Computing Machinery),
DOI: 10.1145/3369740.3373015.
You can read the full text:
Contributors
The following have contributed to this page
