Skip to Content

See and Avoid with a Fuzzy controller optimised using Cross-Entropy method

 

A visual servoing system has been implemented using Fuzzy controller for See and avoid task. This fuzzy controller has three inputs (error, dot error and the integral of the error) and one output, the angular velocity for the orientation of the quadcopter. Each input has a gain which is set by the optimization method of Cross-Entropy (CE) (http://iew3.technion.ac.il/CE/) during the tests made in the simulation ROS-Gazebo (http://www.ros.org/wiki/gazebo). 

The image of the simulated camera onboard the quadcopter at the simulated world in the ROS-Gazebo was send to an external program which process it to obtain the position of the object to avoid and send this information to the Fuzzy controller which generate the a command which it sent to the ROS-Gazebo. A diagram of the execution process is shown next.

 The uses of the Cross-Entropy contribute with a importance reduction in the number of rules from 175 to 35 rules. The optimization of the controller was made using the cost function of Integral of Time and Absolute Error (ITAE) during five seconds test. More than 300 controllers have been tested obtaining a great reduction of the ITAE.

This work has been develop with the collaboration and during the international stay at the ARCAA (Australian Research Centre of Aerospace Automation) of the Queensland University of the Technology (Brisbane-Australia).