Study Of Odor Source Localization Method For Multi-robot Based On Improved SMA

Along with the booming chemical industry,many accidents are occurring,the most important of which are explosions and fires caused by dangerous gas leaks.These accidents seriously threaten social stability and the safety of people.In case of gas leakage,the traditional method would allow rescuers to carry equipment to the accident site,which could easily cause injury to the rescuers.Therefore,the research of robot active olfaction in the direction of odor source localization(OSL)needs to be paid attention.In order to make robotic active olfaction more efficient and practical,this paper proposes a multi-robot OSL method based on the improved slime mould algorithm(SMA)and validates it in simulation environment and embedded system.The main contents of this paper are as follows:(1)Some modifications were made to the original SMA based on the actual OSL scenario.The robots were allowed to start from three regions,namely high concentration region,near odor source low concentration region and far odor source low concentration region,to verify the relationship between the search efficiency of the robots and their starting positions.The experiments were conducted in a simulated experimental environment simulated by Fluent software,and it was found that its search success rate exceeded 90% when the number of robots was greater than 4,but more iterations were needed to finding the plume when the robots started from a place far from odor source.(2)For the problem that SMA requires too many iterations in the plume finding stage,the improved SMA algorithms,namely,the concentration adaptive step slime mould algorithm(CASMA)and the wind direction adaptive step slime mould algorithm(WASMA),are proposed to make full use of the gas concentration information and wind field information in the environment.Through simulation experiments and comparative analysis,it is found that the search success rate of these two algorithms is higher than that of SMA no matter from which region they start from,and its search success rate exceeds 90% when the number of robots is greater than 3,and the number of iterations of search is less.(3)To enable the above algorithm to be used in a real robot hardware environment,a Node MCU multi-node based multi-robot cooperative system is designed,where each node obtains the simulation environment information from the cloud and then performs distributed computation of the CASMA in the embedded hardware.The results show that the above algorithm is suitable for embedded computation,and each node takes only 2-10 ms for each round of position computation,which ensures the real-time operation of robot collaboration.(4)To enable the robot to perform real-time plume confirmation during plume finding,a lightweight algorithm,Quantum Neural Network(QNN),was used for odor recognition.QNN can be regarded as a modified version of back propagation neural network(BPNN),but simple to implementation as a conventional BPNN.Through experiments and comparative analysis,the classification accuracy could increase along with the increase of the number of quantum intervals in QNN(highest accuracy of 99.38 %),and the performance of QNN are superior to BPNN(averaged classification accuracy of 82.30 %),SVM(98.87 %)and k-NN(98.75 %).