Skip to content

Latest commit

 

History

History
83 lines (55 loc) · 6.49 KB

README.md

File metadata and controls

83 lines (55 loc) · 6.49 KB
Raw

Robotic Action and Perception

The current repository contains the code developed for the course Robotic Action and Perception, Masted Degree in Mechatronics Engineering - Department of Industrial Engineering - University of Trento, held by professors De Cecco Mariolino, Luchetti Alessandro. The project is supervised also by Tavernini Matteo, CEO of the startup Robosense

Running the code

To run the SLAM algorithm is necessary to load a set of data and properly configure the config.m file selecting the correct path. Then the main.m can ben executed

Laser Rangefinder Navigation

Data

Data inside the Data folder are obtained by means of the gridmap navigation simulator application, an open source software that simulates the mobile robot motion as well as it mesures (with uncertainties); in particular we can generate the following files:

  • simul_LASER_LASER_SIM.txt: each row contains the measurements of the LIDAR at each time step; in the particular case, the scan has a field of view of 180° that's divided in 361 samples;
  • simul_LASER_LASER_SIM_times.txt: each row contains the absolute acquisition time of the respective LIDAR scan in the previous file;
  • simul_ODO.txt: each line contains the increment of the robot odometry, variables $x,y,\theta$;
  • simul_ODO_times.txt: each line contains the absolute acquisition time of the respective odometry increment.

Repository management

In the Scripts folder there are developed the main script parts that are then imported in the main file:

  • load_data: read laserscan and odometry data and performs some basic preprocessing;
  • plot_raw_data: given the cellarray of laserscans in load_data, it provides some plots to visualize the raw provided data as well as the extracted features;
  • EKF: contains the main loop of the Ektended Kalman Filter

Classes

Different classes are developed to get different levels of abstraction of informations. Naming and ideas mainly obtained by this article (also cited in the references).

Everything has been coded in a Object-Oriented Programming to separate the main function and improve code readibility.

Laserscan

Handles information of a laserscan. It takes as input a the polar measurement and is able to convert them in cartesian space.

Internally embedds an algorithm to extract feature from the cartesian map thus generating a vector of observations for the given data that will be used in the Kalman prediction step as well as in the map update.

Odometry

Stores information of the odometry system on the robot in order to have easier comput.

Robot

Stores current information of the robot, in particular it's state estimate and covariance matrix.

It contains function that allows for a simple calculation of jacobians used in the EKF.

Observation

An observation is regarded as a feature extracted in the robot's reference frame and it's characterized by a cartesian position (w.r.t. the robot) and an uncertainty that's computed based on the one of the LiDAR.

Landmark

A landmark can be regarded as an observation projected into a fixed reference frame; it's described by a cartesian coordinate and it's uncertainty.

The goal of the SLAM is thus to localize the robot based on the landmark that it sees and handling newly seen landmarks.

Map

Handles the structure of the map, storing all landmarks inside it and providing functions aiming at closing the loop and computing the correspondences between current observations and landmarks inside the map itself.

Notes

File load_data.m reads the simulated data and builds cell-arrays of Laserscan and Odometry objectes. While loading the laserscans, it also pre-computes the features to reduce the overhead at EKF runtime. Since this is a highly time-consuming operation, the pre-processed cell-arrays are stored in a folder ProcessedData where they can be cleanly loaded each other time. This can be achieved by enabling the following flag in the main script:

load_precomputed_data = true;

If no pre-processed data are present in the folder, still raw data are read and all features are extracted, saving the computation in the ProcessedData folder. If simulation data are replaced, we can so act in two way: by setting the flag load_precomputed_data = false or by simply deleting the files in ProcessedData.

Bibliography

References for feature extraction:

References for closure loop:

References for EKF:

References for building the map: