The code distributed here provides the implementation of two different algorithms for the solution of object tactile localization: the Scaling Series algorithm and a novel algorithm, the Memory Unscented Particle Filter (MUPF). The MUPF has been tested also on the tactile object recognition, formulated as a localization problem with multiple models. The recognition solution is chosen as that object model who provides the minimum localization error.
Before compiling the code you are required to install
- [YARP](http://www.icub.org, and all the detailed information can be found at http://wiki.icub.org/wiki/Manual#Six._Software.2C_Compiling_YARP_and_iCub)
- CGAL
An example of compilation in Linux is given by:
mkdir build
cd build
ccmake ..
make install
You can run one of the algorithm typing in the command line:
localizer num_of_trials "mupf" --from configuration file
-num_of_trial is the number of times you want to run the algorithm and to have statistics about
-"mupf" string enables the use of MUPF algorithm. Otherwise, Scaling Series is used.
Memory Unscented Particle Filter for 6-DOF Tactile Localization, G. Vezzani, U. Pattacini, G. Battistelli, L. Chisci, L. Natale, submitted to IEEE Transaction on Robotics, 2016, preprint available on arxiv:1607.02757v2
A Novel Bayesian Filtering Approach to Tactile Object Recognition, G. Vezzani, N. Jamali, U. Pattacini, G. Battistelli, L. Chisci, L. Natale, IEEE International Conference on Humanoid Robots, 2016, pp. 250 - 263
