Extending the Occupancy Grid Concept for Low-Cost Sensor-Based SLAM

Abstract

Abstract:The simultaneous localization and mapping problem is approached by using an ultrasound sensor and wheel encoders. To account for the low precision inherent in ultrasound sensors, the occupancy grid notion is extended. The extension takes into consideration with which angle the sensor is pointing, to compensate for the issue that an object is not necessarily detectable from all positions due to deficiencies in how ultrasonic range sensors work. A mixed linear/nonlinear model is derived for future use in Rao-Blackwellized particle smoothing.

Rao-Blackwellized Particle Smoothing for Occupancy-Grid Based SLAM Using Low-Cost Sensors

“…For both cases each grid cell in the map actually consists of eight sub-cells corresponding to different directions in the environment according to the occupancy-grid extension mentioned in Sec. 1, see [Nordh and Berntorp, 2012]; what is visualized is the direction from which it is most likely to observe something, for each cell. In the maps presented the color blue represents areas that are unexplored by the sensor.…”

“…where ξ ∈ R 7x1 . The state vector z ∈ R n contains the cells of the modified occupancy grid, where the size n depends on the map dimension, resolution, and the number of subcells used in every cell in the occupancy grid (for details see [Nordh and Berntorp, 2012]). The measurement function C(•) and measurement y m are parametrized in the nonlinear state vector ξ and the ultrasound range measurement r. The inputs enter in the nonlinear states, and z is linear given ξ.…”

“…In [Nordh and Berntorp, 2012] we performed SLAM using differential-driven mobile robots equipped with an ultrasound sensor. To handle the deficiencies with sonar sensors we extended the occupancy-grid concept by dividing every cell into subcells.…”

“…Further, we developed a conditionally linear Gaussian state-space model for use in SLAM. In this paper we propose a novel measurement model aimed at sonar sensors, and extend the work in [Nordh and Berntorp, 2012] to include Rao-Blackwellized particle smoothing (RBPS) as a means for SLAM. Rao-Blackwellization takes advantage of a linear substructure in the model, which can be handled by a Kalman filter.…”

An improved FastSLAM 2.0 algorithm based on FC&ASD-PSO

“…41 Ultrasound sensors are very sensitive to the angle of an object's surface relative to sensors. 42 In the FastSLAM 2.0 algorithm, the estimations of the robot pose and landmark positions are updated by the difference between the observation z t and the predicted observationẑ t . This updating depends on Q t and R t .…”