Article Preview
TopIntroduction
SLAM problem is a situation where a mobile platform, being placed in an unknown environment, has to build a map and find its location simultaneously. There are many algorithms that are applicable to mobile moving platforms, such as a robot vacuum cleaner, a reconnaissance drone, or even a rover. However, the process of building a map can be accelerated if several agents are used at the same time, where each explores its part of an environment, and in the future, the individual parts can be combined into one picture. The SLAM problem that is being solved by several agents is referred to as multi-agent SLAM problem further.
The paper extends the work (Filatov, 2019), the most important of which questions are described below. The first part of this paper continues the work of (Krinkin, 2017) and presents a brief description of existing multi-agent SLAM algorithms, describes most common approaches and shows their advantages and disadvantages. It describes the approaches how to extend a single-agent SLAM algorithm to the multi-agent case and. It also demonstrates the approaches that were initially based on a multi-agent architecture. Thus, the high-level description of state-of-the-art approaches is presented in this paper as opposed to the previous paper which focused on doing a survey of several algorithms.
The second part of this paper covers the issue of map merging (a problem about combining results of several agents). Each existing algorithm solves this problem in its own way that is most commonly based on the architecture of current approach. The paper presents the high-level solution to this problem.
The third part is an intuitive extension of a laser 2D grid based single-agent SLAM algorithm. The common idea of the suggested algorithm is to fulfill two statements:
- •
Each agent should work individually, which means that there should not be a server that might make the whole system vulnerable to the loss of this agent.
- •
The algorithm should be successfully launched on the low performance hardware without delays and freezing. (This paper considers such hardware to be comparable with Raspberry Pi 3B).
To fulfill the first statement agents should contact each other without an intermediary, i.e. agents should be gathered, for instance, in an ad-hoc network. The second statement makes the restriction that the algorithm should base on simple approaches which leads to tradeoff between the accuracy and speed of performance.
The paper is structured as follows: Section 2 describes of existing multi-agent algorithms; Section 3 provides a description of the developed algorithm; Section 4 deals with its performance and testing on the real recorded datasets.