This paper presents a road-network search route planning algorithm by which multiple autonomous vehicles are able to efficiently visit every road identified in the map in the context of the Chinese postman problem. Since the typical Chinese postman algorithm can be applied solely to a connected road-network in which ground vehicles are involved, it is modified to be used for a general type of roadmap including unconnected roads as well as the operational and physical constraints of UAVs. For

more »

... s, a multi-choice multi-dimensional Knapsack problem is formulated to find an optimal solution minimising a flight time and then solved via mixed integer linear programming. To deal with the dynamic constraints of the UAVs, the Dubins theory is used for path generation. In particular, a circular-circular-circular type of the Dubins path is exploited based on a differential geometry to guarantee that the vehicles follow the road precisely in a densely distributed road environment. Moreover, to overcome the computational burden of the multi-choice multi-dimensional Knapsack algorithm, a nearest insertion and auction based approximation algorithm is newly introduced. The properties and performance of the proposed algorithm are evaluated via numerical simulations operating on a real village map and randomly generated maps with different parameters. International Journal of Systems Science IJSS 2 H. Oh et al other is the Chinese Postman Problem (CPP) finding the shortest path with considering path constraints on an entire network of road. The TSP using multiple UAVs can be considered as a task assignment problem to minimise the cost of time or energy for a certain mission by assigning each target to UAVs, for which binary linear programming (Bellingham et al. (2003)), iterative network flow (Chandler et al. (2002)) and tabu search algorithm (Ryan et al. (1998)) have been proposed. Choset (2001) surveys research results in coverage path planning that determines a path for a robot to pass over all points in its free space. The CPP and its variants (Pearna and Chiub 2005) are normally used for ground vehicle applications such as road maintenance, snow disposal (Perrier et al. (2007) ), boundary coverage (Easton and Burdick (2005) ), and graph searching and sweeping (Alspach (2006) ). In the aforementioned works, since the general vehicle routing algorithms approximate their path to a straight line shape to reduce computational load, the physical constraints imposed on the vehicle are not to be addressed. Recently, Dubins TSP (DTSP) algorithms were developed which can accommodate the physical constraints using the so-called Dubins nonholonomic planar vehicle model constrained to move along paths of bounded curvature without reversing direction. For a single vehicle, Salva et al. (2008) proposed an alternating DTSP algorithm based on the solution to the conventional TSP and on an alternating heuristic to assign the target orientation at each target point. Ny et al. (2012) also developed a DTSP algorithm using a heading discretisation. Rathinam et al. (2007), Ahmadzadeh et al. (2006), Tang and Ozguner (2005) considered the similar problem but using multiple agents. However, these physical constraints have been rarely dealt with in the context of the CPP. Therefore, this paper presents the road-network search route planning algorithms by which multiple airborne platforms are able to efficiently patrol every road segment identified in the map of interest in the context of the CPP. The first part of this paper introduces a conventional CPP algorithm for the case that ground vehicles moves along a connected road-network. Following this, the CPP algorithm is modified to consider a general type of roadmap including unconnected roads as well as operational and physical constraints on speed and minimum turning radius of UAVs. Our previous work (Oh et al. (2011) ) regarding the modified CPP (mCPP) for the road-network search route planning was to formulate Multi-choice Multi-dimensional Knapsack Problem (MMKP) so as to find an optimal solution minimising a path length or a flight time and to solve it via mixed integer linear programming (MILP). The main contributions of this paper are threefold. Firstly, to overcome the computational burden of the MMKP algorithm and to induce a real time solution, a nearest insertion algorithm combining with an auction-based negotiation is newly proposed to be applied for the search route planning by multiple UAVs. Secondly, in order to accommodate the physical constraints of the UAV in the search pattern design, this paper uses the Dubins trajectory (Dubins (1957)) which is the shortest path connecting two configurations represented by position and pose under the constraints of a bound on curvature or turning radius. Although a CSC (circular-straight-circular) type of the Dubins path has been generally used for the path planning of autonomous systems, this paper exploits both a CCC (circular-circular-circular) and a CSC type of the Dubins path to precisely cover a densely distributed road environment. Besides, the detailed mathematical derivation of constructing the Dubins path is presented using the principle of differential geometry. Lastly, this paper systematically investigates the performance of the proposed algorithm depending on different map sizes, path planning methods (straight line and Dubins path) and the number of UAVs by using Monte Carlo simulations. Based on these results, an efficient UAV team size and path planning method is advised for the specific road-network search mission considered in this paper. Furthermore, to clarify the benefit of the proposed algorithm, this paper compares the performance of the MMKP optimisation and the approximation algorithm in terms of computational load and flight time for a specific road-network search scenario. The remainder of this paper is organised as follows. Section 2 defines the problem of roadnetwork search route planning introducing the TSP and the CPP. Then, Section 3 proposes the