When a hazardous contaminant container is accidentally or intentionally released in an open area, the material inside the container will mix with air to become an aerosol and be diffused into the surrounding area with the flow of air. The traditional approach of using an animal such as a dog for detecting, tracking and seeking odor sources cannot be used for locating such highly fatal aerosols. Continued sampling and careful search of the entire suspected area by a professional human could finally locate all of the aerosol emission sources, but such approach is not efficient in terms of time requirements, not to mention the risk to the human operators. 

    With today’s advanced technology, many research groups have investigated sending robots equipped with a gas/aerosol-sensitive system to the suspected area to locate emission sources. In contrast to range-finder sensors such as sonar or laser, which can remotely detect obstacles in a given range, the electric gas sensor installed on the mobile robot can only provide gas concentrate information about a very small area and the gas sensors usually cannot provide an instantaneous and precious measurement of the gas concentration because of the sensor’s long response time and even longer recovery time. However, the natural diffusion phenomena of gases and aerosols tend to spread in the environment inducing a concentration gradient that can be used as a clue for tracing emission sources, although this gradient phenomena are very easily be impacted by the air flow and geographical status of the environment. Currently, most of the research focuses on investigation techniques for locating the sources by using a single robot. Because of the spatial limitations of a single robot, few robotic systems have been developed that demonstrate the capability to carry out the source localization task in a large-scale area with unstructured environment. 

    In this paper, we provide a biasing expansion swarm-based exploring approach (BESE) for helping the robots efficiently explore the environment and find all emission sources as soon as possible. The key concept in this approach is swarm behavior. By applying the three properties of the swarm behavior: separation, cohesion and alignment, our approach can ensure the robot group attains large region coverage, dynamically stable connectivity and fast targets approach. Using an occupancy grid map to represent the unknown environment, Ad-hoc network for wireless communication and a modified particle swarm optimization algorithm (PSO) for path planning, the BESE controls the robot swarm movement in the near optimal orientation for quickly locating the emission sources. The simulation presented in this paper shows the BESE can achieve better performance than the gradient-seek decent approach in an environment that has complex aerosol accumulation, distribution and multiple emission sources.

    For quickly approaching the emission sources, different from the gradient seek approach, which greedily search for highest concentration value at its surround environment and is easily been trapped in local maximal aerosol accumulation, BESE approach simultaneously considering all concentration value that currently been sensed by other robot members in the swarm and find out the positive gradient direction of whole coverage area of the robot swarm. This will make the robot swarm immunity to the random sensor errors, local aerosol accumulation and other local interference effect during their exploring. The BESE was tested in a two aerosol-emission sources localization simulator under different environment scenario. The result shows the BESE have a better performance than the gradient seek approach. At the same time, from the simulation we noticed that the BESE approach controls the whole robot swarm quickly approach the sources compare with gradient approach that controls robot members randomly deployed in the environment. Although in our source searching task scenario, one robot move into the source located cell indicates the task is finished. Gathering all robot swarm at the area near the sources is not required. This “side effect” feature of the BESE approach give us a clue that it is also a potential task allocation algorithm in the multiple heterogeneous agents collaborative applications such as disaster rescue.