Humboldt State University1, Computer Science, Arcata, CA 95521 Monmouth College2, Monmouth, IL 61462 DePauw University3, Computer Science, Greencastle, IN 46135
A prominent focus in artificial intelligence is investigating autonomous vehicles in navigation and transportation. Autonomous navigation systems are commonly orientated toward driving on streets or highways to reach a single final destination, using lane dividers and other indicators to direct them. Due to the lack of research being done using autonomous vehicles for exploration, we decided to examine this idea further. The focus of our research was to develop and test a model for a multi-agent system composed of autonomous vehicles. There are many benefits to incorporating multiple vehicles in a system. For instance, multiple vehicles should be able to investigate a region in less time than is feasible for a single vehicle. Additionally, merging collected data from each vehicle should result in a more complete and accurate representation of a travelled region. Furthermore, using autonomous vehicles to explore unknown regions allows examination of dangerous terrains and unstable structures without placing humans in hazardous situations. To achieve our goal for this model, we concentrated on three key aspects. Computer vision, navigation, and communication are the foundations of our model. Computer vision allows the vehicle to detect potential obstacles contained in the region of exploration. Attention was given to the placement of sensors on a vehicle to provide ample range of vision for obstacle detection and collision prevention. To facilitate navigation, each vehicle is given a set of destinations within a specific section of the overall region. The vehicles attempt to travel to each destination, mapping as they go, until they have reached all possible destinations in their sets. Communication does not exist between the vehicles; however, all vehicles can communicate with a master computer to provide details about their mapped sections, eventually leading to a map of the entire region. In our attempt to implement this model, two approaches were analyzed. The first approach was a software simulation. Utilizing position and heading, it allows simulated vehicles to navigate to destination points generated in the region. Regrettably, due to time constraints, the implementation of multiple vehicles is limited. The second approach was a hardware simulation. The vehicles have the ability to detect and avoid obstacles; they are aware of their heading; and they are capable of communicating to the master computer. Unfortunately, vehicles are currently unable to determine their absolute position. In testing both of these simulations, we have discovered it is often difficult to remain accurate in calculations and in travel; however, each simulation is capable of compensating for the other’s weaknesses. In the future, we would like to integrate these two simulations. This would provide a single working system, allowing us to reach our goal of using multiple autonomous vehicles for exploration of an unknown space. Until then, these simulations provide a solid foundation for future work.
[Abstract (DOCX)]