Bowen Lu

Mobile sensor networks for modelling environmental pollutant distribution

This article proposes to deploy a group of mobile sensor agents to cover a polluted region so that they are able to retrieve the pollutant distribution. The deployed mobile sensor agents are capable of making point observation in the natural environment. There are two approaches to modelling the pollutant distribution proposed in this article. One is a model-based approach where the sensor agents sample environmental pollutant, build up an environmental pollutant model and move towards the region where high density pollutant exists. The modelling technique used is a distributed support vector regression and the motion control technique used is a distributed locational optimising algorithm (centroidal Voronoi tessellation). The other is a model-free approach where the sensor agents sample environmental pollutant and directly move towards the region where high density pollutant exists without building up a model. The motion control technique used is a bacteria chemotaxis behaviour. By combining this behaviour with a flocking behaviour, it is possible to form a spatial distribution matched to the underlying pollutant distribution. Both approaches are simulated and tested with a group of real robots.

Environmental field estimation of mobile sensor networks using support vector regression

This paper presents a distributed algorithm for mobile sensor networks to monitor the environment. With this algorithm, multiple mobile sensor nodes can collectively sample the environmental field and recover the environmental field function via machine learning approaches. The mobile sensor nodes are able to self-organise so that the distribution of mobile sensor nodes matches to the estimated environmental field function. In this way, it is possible to make the next-step sampling more accurate and efficient. The machine learning approach used for function regression is support vector regression (SV R) algorithm. A distributed SV R learning algorithm is used for on-line learning. The self-organised algorithm used for deployment is based on locational optimisation techniques. In particular, Lloyds algorithm for optimising centroidal Voronoi tessellations (CV T) is used to spread mobile sensor nodes over the monitored environment. The environmental field function is simulated in static and dynamic settings and the demonstration on the simulated environments shows the proposed algorithm is effective.

A creative computing approach to 3D robotic simulator for water pollution monitoring

This paper presents a creative computing approach to the development of a 3D robotic simulator that could be used to develop control algorithms for a team of robotic fishes to effectively monitor water pollution. The computational fluid dynamics of a marine environment is taken into consideration, which differs from most existing robotic simulators and shows its creativity. It is the first 3D robotic simulator that enables users to develop control and navigation algorithms for pollution monitoring by a school of biologically inspired robot fishes. Additionally, this simulator has an interface to force fields and provides the diversity of simulation scenarios. The record-and-replay strategy is deployed in the simulator so that the fluid computation is done offline and recorded into a log file, which can then be replayed during a simulation. Experiments were conducted and results show that the proposed approach is feasible and has potential to be developed for a wider range of underwater applications.

Tracking and modeling of spatio-temporal fields with a mobile sensor network

This paper presents an approach to modeling and tracking spatio-temporal field functions by using a mobile sensor network. The modeling tool used is the Gaussian process regression (GPR) technique characterized by a spatial kernel function. Due to the dynamic nature of spatio-temporal fields, the sampled data points have to be selected to remove the outdated data points before they are used for modeling. Less data points also reduces the computational complexity of GPR. The data selection is conducted via an information entropy based selection criteria. With the selected data points and the estimated GPR model, the mobile sensor nodes are controlled to cover the interested region and track the field function. The coverage and tacking control are implemented by using the centroidal Voronoi tessellation (CVT) method with a constraint of limited communication range. The algorithms are verified by using simulation and real robot experiments. The environmental field in the practical experiment is a moving light intensity distribution. The experimental results show the robots are able to model and track the moving field.

Spatial function estimation using Gaussian process with sparse history data in mobile sensor networks

This paper presents a sparse history data based method for modelling a latent function with mobile wireless sensor networks. It contains two main tasks, which are estimating the latent function and optimising the sensor deployment. Gaussian process (GP) is selected as the framework according to its excellent regression performance. History data can improve the modelling performance with small amount of sensors in static or slowly changed environment. However, the GP kernel size is expended. On the one hand, in other kernel based (or non-parametric) methods, computation cost increases fast with kernel size. To control the size of GP kernel, informative vector machine (IVM) is introduced for history data selection. On the other hand, centroidal Voronoi tessellation (CVT), a gradient based method, is adopted for optimising sensor deployment. Simulation results with different data selection methods and analyses are given. Its proved that the data selection is effective in reducing computation cost and keeping the precision of the estimated model.

Distributed least square support vector regression for environmental field estimation

A distributed approach to monitoring the environmental field function with mobile sensor networks is presented in this paper. With this approach, a mobile sensor network is capable to estimate a model of field functions in real-time. This approach consists of two stages, a field function learning stage and a locational optimising stage. A distributed least square support vector regression (LS-SVR) is developed for the field function learning stage. On the locational optimising stage, a gradient based method: centroidal Voronoi tessellation (CVT) is used to allocate each sensor nodes position. These two stages are running alternately in a loop so that the field function learning stage can keep updating the field function with new sensor readings resulted from the locational optimising stage, and simultaneously, the locational optimising stage can relocate sensor nodes according to a more accurate field function model. Eventually, the field function is estimated and the sensor nodes are distributed based on the estimated model. The simulation results given in this paper show the effectiveness of this approach.

Using CFD in robotic simulators for pollution monitoring

This paper presents the RoboShoalSim, a robot simulator that was developed for the EU Robotic fish shoal project to monitor pollution levels in sea ports around the world. The simulator is different from those in use in that it incorporates not only pollution models of the pollutant in the simulated port but uses computational fluid dynamics of the sea port to generate closer to reality pollutant dynamics. In addition, a physic engine is used with the aim of studying the effects of the Robotic fishes dynamics on the pollutant plume structure. In this paper, the general architecture is described, with preliminary examples of utilization.

A behavior based control system for surveillance UAVs

Unmanned Aerial Vehicles (UAVs) is required to carry out duties such as surveillance, reconnaissance, search and rescue and security patrol missions. Autonomous operation of UAVs is a key to the success of these missions. In this chapter, we propose to use a behavior based control architecture to implement autonomous operation for UAV surveillance missions. This control architecture consists of two layers: a low level control layer and a behavior layer. The low level control layer decomposes 3D motion of UAVs into several atomic actions, such as yaw, roll, pitch, altitude, and 2D position control. These atomic actions together serve as a basis for the behavior layer. The behavior layer consists of a number of necessary behaviors used for surveillance missions, including take-off, object tracking, hovering, landing, trajectory following, obstacle avoidance amongst other behaviors. These behaviors can be instantiated individually or collectively to fulfill the required missions issued by human operators. To evaluate the proposed control architecture, the commercially available DraganFlyer QuadRotor was used as the UAV platform. With the aid of an indoor positioning system, several atomic actions and a group of behaviors were developed for the DraganFlyer. Real testing experiments were conducted to demonstrate the feasibility and performance of the proposed system.

Sparse Gaussian Process for Spatial Function Estimation with Mobile Sensor Networks

Gaussian process (GP) is well researched and used in machine learning field. Comparing with artificial neural network (ANN) and support vector regression (SVR), it provides additional covariance information for regression results. By exploiting this feature, an uncertainty based locational optimisation strategy combining with an entropy based data selection method for mobile sensor networks is presented in this paper. Centroidal Voronoi tessellation (CVT) is used as a locational optimisation framework and Informative Vector Machine (IVM) is applied for data selection. Simulations with different locational optimisation criteria are conducted and the results are given, which proved the effectiveness of presented strategy.