Aim
Aim of this project is to develop a multi-robotic systems for the applications in agriculture. The idea is to develop groups of variety of robots where each group is capable of doing some task of agriculture. So, the beginning of this project is from SLAM.
|
IntroductionTo make the grid maps using known poses of a mobile robot, I basically implemented the Bayes Filter and used the log odds form to add the every instantaneous map to update the complete map. Below are some of the calculations which will show how to implement bayes filter and find grid maps using log odds form.
Grid Maps: Grid maps are basically 2D maps with grids of probabilities. The white space in the grid map indicates the empty space while the black color indicates the presence of obstacle. They can be generated by 2d laser scanners or rotating ultrasonic sensor. Probabilistic Robotics: Probabilistic approaches are very important in the scenario of mapping and localization of mobile robots. Probability values are directly stored in the log odds form. Log odds terms are used so that these values for each map generated by sensor can be simply added to the log odds term of the maps which has unknown environment. Probabilistic robotics is applied in sensor fusion algorithms like Kalman Filter, EKF, UKF and particle filters. They are also used for localization of robot using probabilistic approaches. For ex. If the robot is given an action command ( u ) then, robot will traverse certain distance and will cover up certain angular motion as per the odometry. But the odometry results won't be perfect. So, using the motion models we can predict the possible positions of the robot and using one sensor data ( like GPS or imu sensor readings ) may allow us to localize the robot. Completed Work: Study of grid mapping with known poses, motion models, sensor models, Kalman Filter, EKFs, UKF & Particle filters are studied. Along with the study, all of them are implemented on the V-REP. Problems: Simulation has been done on V-REP. Even if V-REP has a very good API still, there is a communication lag between them and also the data published by hokuyo sensor is a 3x684 matrix after every time step. This creates issues in V-REP and MATLAB. So the solution is to use Gazebo and separate nodes to subscribe and publish the data. Future Work:
|
Literature StudyResultsTo study more about mapping, slam, filters ... click here
|