Odor Source Localization Based On A Quadrotor In Large-scale Outdoor Environment

This paper presents a method for locating odor sources using a quadrotor and a TDLAS gas sensors in large-scale outdoor environments.This method can be applied to the routing inspection,searching and localization of toxic/harmful gas lakage or chemical hazards in outdoor three-dimensional search space.In this paper,chemical plume tracking and source’s location estimation in three-dimensional environment are studied,and the following work is carried out.This project first establishes an experimental platform for chemical plume tracking in three-dimensional search space.The experimental platform uses quadrotor,and the Pixhawk2 flight control panel on which it is mounted is redeveloped.Through the u ORB process communication mechanism built in the flight control panel and MAVLink communication protocol,the PAYLOAD data format under the framework of MAVLink protocol is customized,and the flight parameters of the quadrotor are obtained in real time through wireless data transmission module,which lays the foundation for chemical plume tracking in three-dimensional search space.Secondly,the olfactory robot searches for odor sources by detecting odor concentration and wind speed/direction information,but for quadrotor,the anemometer is too heavy to be airborne,and the downwash airflow generated by the rotator of the UAV strongly disturbs the local wind field.It is difficult to obtain accurate wind speed/direction information using the anemometer.Therefore,real-time estimation of wind speed/direction(wind vector)without using an anemometer is especially necessary.The wind vector estimation method based on the unmanned aerial dynamics model is a feasible method,but it has some limitations.Two important parameters(the unmanned aerial vehicle thrust coefficient and the resistance factor)are unknown,and the relationship between the rotator motor speed and the equivalent voltage corresponding to the PWM input signal is complex.This paper presents a real-time estimation method of wind vector based on artificial neural network.By setting up a gray box model for the calculation of wind vector based on artificial neural network,the determination of the thrust and air resistance coefficients of quadrotor can be avoided,and the relationship between the rotation speed of the rotator motor and the equivalent voltage of the PWM input signal can be avoided.This overcomes the shortcomings of existing wind vector estimation methods based on unmanned aerial dynamic models.By analyzing the unmanned aerial dynamics model,the quadrotor unmanned aerial vehicle rotator voltage and(calculated from the weight,acceleration and attitude angle of the unmanned aerial vehicle)are selected as the input of the neural network,and(which is a function of the ground speed,attitude angle and wind speed of the quadrotor)is selected as the output of the artificial neural network.The ground speed,acceleration,attitude angle and wind speed/direction information of the quadrotor are obtained through calibration experiments.The neural network is trained,and the real-time estimation model of the wind vector artificial neural network is established.The real-time estimation method of wind vector proposed by BP network and RBF radial basis function network is implemented and validated in the experiment.The results show that the root mean square error of the two real-time estimation models is less than 0.02m/s,and they both have strong applicability and stability.Finally,a set of smoke plume tracking and odor source location estimation methods for quadrotor based on TDLAS rotating scan mode is presented.In order to improve the efficiency of searching for odor plumes in three-dimensional outdoor environments,on the one hand,the TDLAS gas detector on the quadrotor rotates continuously downward at a certain deflection angle to increase the detection space range;on the other hand,a ring sequence of odor concentration integral values is constructed and updated synchronously during the TDLAS rotation scan.In the odor search process,the maximum value and its corresponding direction angle were found in the odor concentration integral ring sequence in each control cycle,and the next search direction of the quadrotor was planned out,which gradually approached the odor source.During the task,the location of the quadrotor is determined by real-time convergence,and the location of the ground odor source is estimated and the odor source search process is terminated.The simulation results show that the smoke plume tracking method proposed in this paper based on TDLAS rotary scan mode has high search efficiency and success rate,and the odor source location estimation method is effective and accurate.