SLAM Constructor for ROS
Simultaneous Localization and Mapping (SLAM) methods are essential for mobile robots which are supposed to act in an unknown environment. In spite of various algorithms have already been proposed, an algorithm that robustly solves the problem in general case and satisfies performance constraints is still a subject of research. Unfortunately, there is no publicly available framework that provides a common set of components in order to speed up SLAM research (frameworks and toolkits that simplify development of particular SLAM parts are not taken into account).
- creation of a framework that acts as a constructor of SLAM algorithms (a researcher is supposed to connect available components by himself and add necessary modifications);
- implementation of full set of basic components that can be assembled into the most common (fundamental) SLAM algorithms;
- creation of the infrastructure and tools for SLAM algorithms debugging and analysis;
- creation of service for management SLAM datasets (converting, storing, providing etc);
All software we develop is supposed to be executed in Linux and Robot Operating System environment. (We did preliminary research wide spectrum of existing tools and environments like MTK, MRPT and others and have seen that ROS is most promising choice)
- Support components for graph-based SLAM methods;
- Implementation extra scan matchers for Laser based SLAMs
- Support extended set of sensors and measurements (3D scans, monocular/stereo cameras)
- Vergent stereo vision implementation
- SLAM datasets service
- ROS SLAM Testing Farm: (virtualized, container based environment, for semi-autonomous SLAM algorithms testing)
Proceedings of the 24st Conference of Open Innovations Association FRUCT,
Modern SLAM (Simultaneous Localization and Mapping) algorithms launched on a moving agent are bounded with its computation resources. The consistent way out is to add more computing agents that might explore the environment quicker than one and thus to decrease the load of each agent. This paper presents the state of art in area of Multi-agent SLAM algorithms and describes problems that are faced in front of a developer of such approach. The outstanding problem of Multiagent SLAM - merging of maps built by separate agent during algorithm is also considered in this paper. Moreover the algorithm that extends laser 2D single hypothesis SLAM for multiple agents is introduced with evaluation of its performance.
An autonomous self driving platform receives information about environment using only its onboard sensors. And it seems obvious that using several sensors could provide more certain information with reduced measurement error. But a general question is how to fuse measurements from different kinds of sensors (like a camera and an accelerometer) to get refined data about a platform or world state. This paper presents a theory based on groups that proves a possibility of correctness of error extraction from a moving model. And there are results of application this theory on fusing measurements from two sensors: odometer and scan matcher