Toward Optimization of AGV Path Planning: An RRT^*-ACO Algorithm

Abstract

Automated guided vehicle (AGV) smart parking provides a new solution to solve the problem of urban parking difficulties. In the AGV parking lot, whether the AGV running path is reasonable affects the transportation efficiency of the entire parking lot. RRT algorithm and ant colony algorithm can achieve good path planning effect for AGV. However, the use of uniformly distributed initial pheromones can easily lead to blind search and local optimization in the early stage of the algorithm, resulting in a decrease in AGV efficiency. In this paper, an AGV based on the fast-scaling random treeant colony algorithm (RRT*-ACO) is proposed. The path planning method uses the fast search mechanism of RRT* algorithm and the positive feedback advantage of ant colony algorithm to combine the two algorithms and add them to the RRT* algorithm. The two-way search, heuristic dynamic sampling and dynamic step size strategies are used to accelerate the search speed of the RRT* algorithm, quickly generate an initial path, and then deploy different concentrations of initial pheromones according to the initial path to guide the ant colony algorithm search. Then, the silo excavation search strategy is added to the ant colony algorithm, which improves the pheromone update rule and reduces the number of iterations of the ant colony algorithm. Finally, several simulation experiments are carried out on the algorithm in maps of different scales, and the comparative experimental results show that the RRT*-ACO algorithm has fast convergence speed, better global search ability, and the path planning efficiency is significantly higher than that of traditional ant colony algorithm and ant colony system, and has better robustness in large-scale maps.

Recent advances in Rapidly-exploring random tree: A review

“…The algorithm is validated in real time on an HP Z440 workstation equipped with 12 Intel Xeon processors for a long tunnel over 20 m, long and a sloped planar space over 30 m, and an indoor environment containing columns and doors over 40 m, with path generation taking 0.25 s, 0.55 s, and 0.45 s, respectively [ 46 ]. Studies that are relevant to branching strategy improvement and also claim to be large-scale scenarios include 131.4 s to explore an 8 m*8 m map [ 187 ], 56.75 s to explore 143.13 square meters of 182-square-meter indoor maze map with 81.46 % precision and 92.89 % accuracy [ 188 ], 1758.8 s to explore 30 m*30 m map [ 189 ], 555.6 s to explore 1000 m*1000 m with 3D terrain map [ 64 ], 10 s to explore 467*785 pixel maps of complex ocean environment with multiple vortices [ 136 ], 12 machine cycles for power inspection [ 165 ], 53.1 s to explore 20*20 grid maps [ 190 ], 6.654 s, 8.8845 s, 6.654 s, 16.1148 s, 7.7544 s, 6.0529 s to explore the dataset of Chem97ZtZ, gemat12, bcsstk33, kron_g500-logn162, CoAuthorsCiteseer [ 191 ]. The reason for the large gap in time cost is that some results only consider the time required for path generation, while others consider the time required for the robot to explore the map, and the above results demonstrate that branching strategy improvements can handle the challenges of large-scale environments.…”