Odor Source Localization Of Multi-Robot Based On Particle Swarm Optimization In Complicated Environments

In nature, odor information is widely used by various creatures to seek for mates,foods, or exchange information. With the development of sensor techniques, robotics,and bionics, researchers have attempted to use robots with olfaction to build the mapof odor source, localize odor source and distinguish different odor sources since the1990s, which is called active robot olfaction. Part representative applications includeharmful or toxic gas detection, rescue after disasters, explosives or narcoticslocalization, and so on.Considering the complexity of actual environments, this dissertation investigatestracing and localization of odor sources based on modified particle swarmoptimization for different plume environments, including the constraint of limitedcommunication among robots, the noise odor concentration detected by sensors,synchronously localization of multiple odor sources and timely tracing of odorsources in dynamic environments with changing wind.Firstly, considering the constraint of limited communication among robots, amethod of localizing odor sources using multiple robots based on particle swarmoptimization is presented on the condition of abstracting each robot as a particle. Inthis method, a strategy incorporating with a repulsive function is utilized to guide arobot to rapidly search for a plume. Then the range of communication among robots isestimated based on the log-distance loss model of wireless signal propagation to forma dynamic topology structure of a particle swarm and to determine the globaloptimum of particles. Finally, the sampling/recovery time of a sensor is incorporatedto update a particle so as to trace the plume.Secondly, the problem of odor source localization in noise environment isfocused on, and a cooperative search method of multi-robot based on particle swarmoptimization is presented. In this method, a robot is defined as a particle, odorconcentration detected by sensors of this robot is regarded as the fitness of thisparticle, and all robots form the swarm of PSO. By estimating the noise degree ofodor concentration detected by sensors using a dynamical statistic method, animproved bare-bones PSO with interval fitness is proposed to lead the particles searchcooperatively odor source. Then a probability domination relationship suitable tointerval fitness is defined to compare particles and update the local and globaloptimum of particles. Moreover, a Gauss sampling method based on the local andglobal optimum of particles is used to update the positions of particles to localize odor sources.Thirdly, aiming at the problem of multiple odor sources localization, amulti-robot cooperation method based on niching particle swarm optimization isproposed. In this method, a robot capturing the plume and its neighbour form a niche,and different niches are employed to localize different odor sources synchronously.The position of a particle in a niche is updated using an improved PSO consideringthe constraints of the sampling/recovery time of a sensor and the velocity limit of arobot. In order to localize more odor sources, the size of each niche is dynamicallyadjusted based on the aggregation degree of its elements. Based on the similarity ofoptimal particles found by niches, a niche merging strategy is proposed to preventparticles repeatedly searching for the same region. Finally, the position of an odorsource is localized based on the concentration value and the position of a robot.Fourthly, aiming at the problem of odor source localization in dynamicenvironments with changing wind, a method of localizing odor source using multiplerobots based on particle swarm optimization and support vector regression is proposed.In this method, a model predicting concentration of an odor at a location based onsupport vector regression is developed, which takes a robot’s current position as itsinput, and the corresponding concentration value measured by the robot as its output.Then, an improved particle swarm optimization is used to localize odor source, andthe position corresponding to the maximal concentration value obtained by theprediction model is taken as the particle’s global optimum in the observation windowof the robot with the maximal concentration value. In addition, the current position ofa robot is taken as the particle’s local optimum. The velocity and position of a particleis updated based on the above global and local optima. Finally, the position of an odorsource is localized based on the concentration value and the position of a robot.Finally, a global robot path planning approach to evade localized danger odorsources based on multi-objective particle swarm optimization is presented in thispaper. In this method, based on the environment map of a mobile robot described witha series of horizontal and vertical lines, an optimization model of the above problemincluding two indices, i.e. the length and the danger degree of a path, is established.Then, an improved multi-objective particle swarm optimization algorithm withself-adaptive mutation operation based on the degree of a path blocked by obstaclesand an infeasible solutions archive is developed to improve the diversity andfeasibility of a new path. Moreover, a constrained Pareto domination based on thedegree of a path blocked by obstacles is employed to update local leaders of a particle and the feasible and infeasible archives.The proposed methods are applied to odor sources localization using multiplerobots and a global path planning containing danger odor sources to be evaded invarious scenarios and the simulation experimental results confirm its feasibility andefficiency.