UNAM, BioRobotics Lab.
Last updated: Oct. 17, 2022
SLAM using Hidden Markov Models
Member
- Jesus Savage, Head Researcher, UNAM
- Marco Negrete, Team Leader, UNAM
- Oscar Fuentes, PHD Student, UNAM
Link
Abstract
Methods of simultaneous localization and mapping (SLAM) are a solution for the navigation problem of service robots. We present a graph SLAM system based on Hidden Markov Models (HMM) where the sensor readings are represented with different symbols using a number of clustering techniques; then, the symbols are fused as a single prediction, to improve the accuracy rate, using a Dual HMM. Our system’s versatility allows to work with different types of sensors or fusion of sensors, and to implement, either active or passive, graph SLAM. A graph-SLAM approach proposed by the International’s Karto Robotics in Cartographer, the nodes represent the pose of the robot and the edges the constraints between them. Nodes are usually defined according to contiguous nodes except when loop closures are detected where constraints for non-contiguous nodes are introduced, which corrects the whole graph. Detecting loop closure is not trivial; in the ROS implementation, scan matching is performed by Sparse Pose Adjustment (SPA). Cartographer uses an occupancy map in order to estimate the position where the map representation is done via Gmapping. The Toyota HSR (Human Support Robot) robot was used to generate the data set in both real and simulated competition environments. In our SLAM representation, we have wheel odometry estimate according to initial position of the robot, a Hokuyo 2D Lidar scan for observations, and a signal control and a world representation is estimated. We tested our system in the kidnapped robot problem by training a representation, improving it online, and, finally, solving the SLAM problem.
Reference
- Fuentes Oscar , Savage Jesus, Contreras Angel, Vol 21 No 1 (2022): Informatics and Automation. [Link]