Improvement of an adaptive fuzzy-based obstacle avoidance algorithm using virtual and real kinect sensors

The article presents improvements to an already existing fuzzy-based navigational system that was already described in a previous paper. To determine the movement direction of the robot, that system used a rule set that contained 16 rules based on 3 distance measurements [1], but this system was not driving the robot in a straight line when navigating in narrow corridors. To get rid of this error, the fuzzy deduction systems were redesigned and new adaptive field of view and new throttle control were introduced; the results of these improvements are that the robot now moves faster in obstacle-free environments and follows a straighter path in narrow corridors. The developed system uses 3 distance measurements as its input to generate the field of view and determine the movement direction for the robot and generate the reference signal for the expected speed of the robot. To calculate these values, three Mamdani-based fuzzy controllers are used that contain 7, 18 and 3 rules, respectively. Thanks to the new system design, the cumulative error of the sensors and the actuators is decreased, because the vehicle makes less turns when it is between straight walls; and since the curve of these turns are smaller as well, these errors will be smaller than with the previous system. For this reason, the estimation of the robot´s location and the creation of the environmental map (SLAM) are more accurate as well. To compare the different navigational methods in an impartial and unbiased way, there was a need to develop a robot and a Kinect sensor simulator, the main development steps of those are described as well. The final system was tested on the simulator and in a real-life environment as well, and the tests showed that the robot can successfully perform the obstacle avoidance and other navigational tasks as well. At the end of the article, the results of the various implemented sub-tasks (e.g. the adaptive field of view and the throttle control) are compared and this sho- s that the final algorithm works ideally in real-world circumstances; this is mainly noticeable when driving through narrow corridors.