An improved heuristic mechanism ant colony optimization algorithm for solving path planning

Highlights

• A mathematical model of global path planning problem is established.

• A novel variant of ACO for solving global path planning is proposed and named IHMACO.

• The proposed IHMACO has four novel mechanisms.

• The IHMACO is compared with 15 existing approaches for solving path planning.

• Experimental results indicate the superiority of the proposed IHMACO.

Abstract

With the development of artificial intelligence algorithms, researchers are attracted to intelligent path planning due to its broad applications and potential development. The ant colony optimization (ACO) algorithm is one of the most widely used methods to solve path planning. However, the traditional ACO has some shortcomings such as low search efficiency, easy stagnation, etc. In this study, a novel variant of ACO named improved heuristic mechanism ACO (IHMACO) is proposed. The IHMACO contains four improved mechanisms including adaptive pheromone concentration setting, heuristic mechanism with directional judgment, improved pseudo-random transfer strategy, and dynamic adjustment of the pheromone evaporation rate. In detail, the adaptive pheromone concentration setting and heuristic mechanism with directional judgment are presented to enhance the purposiveness and reduce turn times of planned path. The improved pseudo-random transfer strategy and dynamic adjustment of the pheromone evaporation rate are introduced to enhance search efficiency and global search ability, further avoiding falling into local optimum. Subsequently, a series of experiments are conducted to test effectiveness of the four mechanisms and verify the performance of the presented IHMACO. Compared with 15 existing approaches for solving path planning, including nine variants of ACO and six commonly used deterministic search algorithms. The experimental results indicate that the relative improvement percentages of the proposed IHMACO in terms of the path turn times are 33.33%, 83.33%, 35.29%, 38.46%, and 38.46% respectively, demonstrating the superiority of IHMACO in terms of the availability and high-efficiency.

Introduction

Global path planning is to create an ordered viable route from the start node to the target node under the known obstructed environment, and it must ensure that the planned path does not collide with any obstacles. Intelligent path planning is a significant tool for many fields, such as robot path planning [1], unmanned combat vehicles (UCV) [2], vehicle routing problem (VRP) [3], [4], transportation system navigation [5], military command systems [6], cruise missile trajectory planning [7], unmanned aerial vehicle (UAV) trajectory planning [8], [9], fire hazard escape [10], automated guided vehicle (AGV) [11], etc. Due to its wide application, path planning problems have gained lots of attention of numerous researchers, and a large number of optimization algorithms are used to solve path planning problems.

Traditional path planning algorithms mainly include Best-First Search algorithm [12], Dijkstra algorithm [13], A* algorithm [14], Jump Point Search algorithm [15], Breadth-First Search algorithm  [16], Trace algorithm [17], Rapidly-Exploring Random tree (RRT) algorithm [18], Probabilistic Roadmaps (PRM) algorithm [19], etc. Kurdi et al. applied Dijkstra algorithm to the problem of autonomous vehicles [20]. However, Dijkstra algorithm is inefficient due to the large number of nodes traversed, and it cannot handle problems with negative edges. Luo et al. presented an extended Dijkstra algorithm by utilizing the Delaunay triangulation to solve surface optimal path planning problems  [21]. Best-First Search algorithm is similar to the process of Dijkstra algorithm, with the difference that it uses a heuristic function to quickly direct the target node [22]. El Baz et al. proposed an improved parallel Best-First Search algorithm to solve multi-core processor planning problems [23]. Breuker et al. combined base-twin algorithm with Best-First Search algorithm to solve the graph history interaction (GHI) problem  [24]. A* algorithm has excellent performance in path planning problems such as good robustness, and fast response to environmental information [25]. Sang et al. introduced an artificial potential field to improve A* algorithm for unmanned surface vehicle formations [26]. Li et al. applied bidirectional A* for robot path planning  [27]. However, A* algorithm is not well suited for high-dimensional space, and it will lead to the abrupt increment of computational work in a complex environment. By improving the A* algorithm, JPS algorithm was introduced to avoid expanding many useless nodes [28], [29]. Breadth-First Search algorithm uses a carpet-based cascading search strategy to traverse the nodes along the width of the tree. Li et al. employed Breadth-First Search to propose a fast path planning algorithm [30]. RRT and PRM are graph search-based methods for solving dynamic path planning [31], [32]. Breadth-First Search algorithm is a blind search method, it cannot ensure finding the optimal route [33]. Thus, Tripathy et al. presented an improved Breadth-First Search based identification method for solving mobile robot navigation [34].

In recent years, many heuristic bionics algorithms are broadly applied to solve path planning problems, such as particle swarm optimization (PSO) [35], genetic algorithm (GA) [36], ant colony optimization (ACO) [37], etc. However, the traditional heuristic algorithm has limitation in solving path planning problems. For overcoming the inefficient operation disadvantage of GA algorithm [38], an improved genetic algorithm was presented by Flores-Caballero et al. to solve path planning of UAV [39]. Cheng et al. proposed a novel variant of GA for reconfigurable tiling robots [40]. Chong et al. presented an improved adaptive genetic algorithm (IAGA) to solve the problem of path planning for autonomous underwater vehicles [41]. Besides, the PSO is susceptible to getting into local optimum [42]. According to a proposed variant of PSO, Wang et al. introduced a method of quantum PSO for offline path planning in AUV [43]. To obtain high optimal accuracy, Zhao et al. proposed a hybrid PSO for solving serial manipulator time-jerk optimal trajectory planning [44]. Kathen et al. proposed a modified PSO to solve the path planning of advanced safety vehicles [45]. Moreover, with the increasing complexity of path planning problems, many scholars have proposed a variety of improved optimization algorithms. Wu et al. introduced a deep reinforcement learning technique called autonomous navigation and obstacle avoidance (ANOA) to solve the problem of autonomous navigation and obstacle avoidance of the unmanned surface vehicle (USV) [46]. The SA is a probability-based algorithm derived from the principle of solid annealing. Xiao et al. presented an improved SA to realize UAV coverage path planning, and the results verified the high-quality effect of the improved SA [47]. Xiong et al. proposed a distinctive sample-based path planning algorithm by combing inspirations from rapidly-exploring random tree star (RRT*) with a tournament selection method and informative heuristics [48].

Compared with other heuristic bionics algorithms, the ACO has the superiority of parallelism, self-organization, positive feedback ability, etc. However, the traditional ACO contains some disadvantages, such as low efficiency, slow convergence, and easy stagnation. Therefore, many researchers delivered a lot of improvement measures to solve problems regarding path search strategy and pheromone update. Ma et al. proposed a fireworks ant colony hybrid algorithm and new mathematical modeling was established by considering the navigation distance cost and energy consumption cost [49]. Ajeil et al. presented an aging-based ACO, which is realized by combining with grid method to solve path planning [50]. Liu et al. proposed an improved ACO which integrates pheromone diffusion and geometric local optimization for mobile robot path planning [51]. An adaptive potential field ACO was presented by Zhu et al. for lunar robot path planning [52], a heuristic factor and an adaptive state transition were introduced to improve the search ability and convergence speed. Yi et al. improved ACO by introducing a multi-objective programming model to optimize pheromone matrix of multi-automatic guided vehicles [53]. In order to solve indoor mobile robots path planning, Miao et al. presented an improved ACO (IACO) by enhancing pheromone update strategy and state transition probability [54]. An adaptive polymorphic ACO was presented by Jiao et al. for solving path planning of smart wheelchairs, it mainly improved the pheromone update strategy [55]. Tao et al. presented an improved ACO that combines the distribution of initial pheromone, modified heuristic function, and pheromone update rule to solve mobile service robots path planning [56]. To improve the drawbacks of ACO in a large amount of calculation, a new weighted adjacency matrix was proposed by Wang et al. for path planning [57]. Zhao et al. presented an improved ACO by introducing the pheromone updating strategy and the path selection strategy for solving omnidirectional mobile vehicle path planning [58]. Even though the ACO and its variants can produce satisfactory solutions in solving path planning problems, with regard to effectiveness and efficiency, there is still space for further improvement of ACO performance.

In this study, a modified ACO is presented to solve path planning. To overcome the disadvantages of classical ACO, four novel mechanisms are presented, which are adaptive pheromone concentration setting, heuristic mechanism with directional judgment, improved pseudo-random transfer strategy, and dynamic adjustment of the pheromone evaporation rate. The adaptive pheromone concentration setting is presented to strengthen the guiding ability for which optional area is prior selected in the initial searching process. A heuristic mechanism with directional judgment is proposed to heighten the search purpose of algorithm and the smoothness of the planning path. An improved pseudo-random transfer strategy is introduced to heighten the search efficiency and avoid getting into the local optimum. A dynamic adjustment of the pheromone evaporation rate method is presented to increase swarm diversity and enhance global search ability. By combining the four proposed mechanisms with traditional ACO, a new variant of ACO named IHMACO (improved heuristic mechanism ant colony optimization) algorithm is proposed. To verify the effectiveness and superiority of IHMACO, a set of experiments is conducted by comparing it with other path planning algorithms in different instances. The results verify the dominance of IHMACO with regard to the convergence speed and optimal solution search ability.

The remaining parts are structured as follows: Section 2 gives the descriptions to build mathematical model of global path planning. Then, traditional ACO and four novel mechanisms included in the IHMACO are introduced in Section 3. Experiments and analysis are conducted in Section 4. Finally, Section 5 gives conclusions.