Multi-agent Planning In Collision Avoidance Applications

As ship becoming faster, larger, more special, automatic and intelligent, the researches in the field of navigation are paying more attentions on how to insure the sailing security, increase the economic benefits, decrease the cost of fuel and reduce the personnel on board. Furthermore, ship collision avoidance is a hot topic in the ship automatization research. With the development of intelligent information management, more and more new technologies, ideas and methods can be used in this area.After many years researching on ship collision avoidance, many works have been done and many methods have been used. Generally, the research work can be divided into two aspects: the theory of ship collision avoidance and the automatic decision making for ship collision avoidance. However, the research results haven't become systematism yet. So far, there is no one automatic decision supporting system for ship collision avoidance can be used in practical navigation. Most researches based on the self-regard purpose to make decisions for collision avoidance from the point of view own ship. The information exchanging among ships are also been neglected. So the collision avoidance decisions made by each ship may be conflictive and inconsistent. It can't be accepted by all the other ships. And it also can't resolve current problem very well. During the process of executing the collision avoidance plans, even more dangerous situation may happen. Therefore, it's more reasonable and practicable to make collision avoidance decisions based on union-benefit propose and take the information-exchanging of ships into consideration. The union is consisted by several ships with some common goals.In this paper, the Intelligent Agent technology is used, which has been deeply researched in the field of sociology, artificial intelligence, software project and so on. Based on the developed multi-agent decision support system for ship collision avoidance (SCA-MADSS), the multi-agent planning is studied. By analyzing the process of ship collision avoidance carefully, the description format of ship collision avoidance plan is designed and the solution space for plan searching is settled. According to experiences of the navigation experts, some criterions (preference functions) are also built to evaluate the collision avoidance plan for the plan ordering and selecting. Then three multi-agent planning algorithms are created and programmed, they are the distributed planning for self-regard purpose, the centralized planning for union-benefit purpose and the negotiated planning for mutual-benefit purpose. In addition, we discussed the advantages and disadvantages of each algorithm. These algorithms are performed on the platform SCA-MADSS. Through several running examples, some comparisons are made on the problem solving capability and working efficiency among these three algorithms. The example results show that it's not only necessary but also feasible to use multi-agent planning in collision avoidance decision making. Since multi-agent planning pursues mutual-benefit purpose, the obtained collision avoidance plans have lesser conflicts than these given out by traditional experts, and can be accepted by all most ships. Multi-agent planning will have more practicability.