In this paper, we are focusing on the obstacleavoidance technique in the dynamic environmentsuch as manufacturing industry environment. For thisobstacle avoidance technique in dynamicenvironment, new algorithms are designed for mobilerobot navigation. The algorithms are designed byusing two techniques. First one is path planning bypotential field approach, which is one of the mostpopular for path planning due to its less calculationtime, and the second technique is Monte CarloLocalization (MCL) (Fox Dieter et al., 1999), whichis one of the most useful localization techniques formobile robotics. In addition, we designed userfriendly Path planning for Mobile robot in ADynamic Environment (PMADE) V 2.0 software forthe user-robot interaction.Path planning is the art of deciding which route totake for navigation under dynamic environment. Thepath planning is based on the current internalrepresentation of the terrain.It involves manycomputations for continuous movement andsequences while moving between the start and endgoal. The path planning involves the computation of collision free path between the two known goals. Theoverhead global color camera has been used forimage processing. The MCL algorithms use sonarsensors for proper localization of the robot in anunknown dynamic environment. The gradient pathplanning algorithm is introduced in path planningmodule to avoid obstacles and to increase the safetyfactor of the robot. The MCL algorithms use all theposition data continuously from the path planningalgorithms for mobile robot localization.This paper is organized as follows. First, the pathplanning module using potential field method isdescribed. The sub modules of path planningalgorithms global path planning and local pathplanning, is presented. The robot control module ispresented in which the Monte Carlo localizationapproach, gradient path planning method ispresented. In the last portion of the paper, design of the obstacle avoidance algorithms described withGUI. Finally, the paper is concluded with real robotand simulator experiments with its results andanalysis.
Path planning module for mobile robotusing potential field technique
In this method, the real valued potential field isconstructed around the robot (
)) and the potentialfield is kept zero value at a goal position. So that, therobot tries to move toward the lower potential fieldi.e. toward the goal. The goal acts as an attractiveforce on the robot and the obstacles act as peaks, orrepulsive forces. The superposition of all forces isapplied to the robot, which, in most cases, is assumedto be a point in the configuration space. Such anartificial potential field smoothly guides the robottoward the goal while simultaneously avoidingknown obstacles.The basic idea behind all potential fieldapproaches is that the robot is attracted toward thegoal, while being repulsed by the obstacles that areknown in advance. If new obstacles appear duringrobot motion, one could update the potential field inorder to integrate this new information. If we assumedifferentiable potential fields function
) aroundthe robot, we can find the related artificial force
)acting near the obstacles position
denotes the gradient vector
(2)Path planning using potential field algorithms aredesigned by the combination of Astar search method,global static path planning module, local dynamicpath planning module and image processing module.
Pioneer robot control module
The control program includes Monte Carlolocalization algorithms and gradient path planningalgorithms. The sonar sensors are used for robotlocalization and obstacle detection in the dynamicenvironment. The gradient path-planning moduleruns parallel with MCL algorithms. In case the sonarsensors detecting any obstacle, at that instant therobot plans its own path and moves to the next goalby avoiding the obstacle.
3.1. Monte Carlo Localization
MCL algorithms (Fox Dieter et al. 1999) particularlykeeps robot updating its steps continuously in theworkspace. The sonar and motor encoder sensors of the robot send continuous data to the robot clientprogram, which is on the PC. The complete system isclosed loop system. MCL algorithms address theproblem of state estimation from sensor data. MCL isa probabilistic algorithm and maintains a probabilitydistribution over the space.The general idea for MCL algorithms is given as:
Initialize the set of current samples so thattheir locations are evenly distributed andtheir importance weights are equal.
Repeat until done with the current set of samples.
Move the robot a fixed distance and thentake a sensor reading.
Update the location of each of the samples(using the movement model).
Assign the importance weights of eachsample to the likelihood of that sensorreading given that new location (using thesensor model).