Source code is available at:
http://www.cs.cmu.edu/~coral/projects/localization/cgr/source/
The Corrective Gradient Refinement
(CGR) algorithm for Monte Carlo Localization (MCL) uses the state space gradients of the
observation model to improve accuracy while maintaining low
computational requirements. See:
Source code is available at:
http://www.cs.cmu.edu/~coral/projects/localization/source.html
Fast Sampling Plane Filtering (FSPF) is a RANSAC based algorithm for
extracting 3D points corresponding to planar features, given a depth
image. The plane filtered points may be used for localization, or to build polygon maps
of environments.
See:
Cobot uses its Kinect sensor for localization as well as for obstacle avoidance. The Kinect localization algorithm is based on CGR, and runs in real time at full frame rates and at full resolution (640x480 @30fps) while consuming <20% CPU on a single core of the Intel Core i5 540M (2.53GHz) processor. The mean localization error of the robot over experiment trials (of length >4km) while using the Kinect for localization is <20cm and <0.5°.
See:
Depth Camera based Localization and Navigation for Indoor Mobile
Robots, Joydeep Biswas and Manuela Veloso, presented at the RGB-D
Workshop at RSS 2011.
The video below demonstrates real-time localization using the Kinect
sensor on Cobot.
We developed a WiFi based localization system which localized cobot on a graph based map including WiFi data (Means and Standard Deviations) at every vertex of the graph. Cobot 1 succesfully localized and navigated autonomously along this graph based map. For a detailed description of the WiFi based localization algorithm, see: