Efficient Path Planning and Collision Avoidance for Autonomous Robot Navigation

The motion planning problem is a fundamental challenge in the field of robotics. This study presents a novel sampling-based pathfinding algorithm that enables robot to navigate complicated terrain by providing obstacle avoidance and local trajectory planning. Sampling-based pathfinding algorithms have been extensively utilized in motion planning challenges. The method has several limitations, including a sluggish convergence rate, significant variation in search time, a vast and dense sample space, and non-smooth search routes. This study introduces RE-RRT* (Robust and Efficient RRT*), an enhanced version of the sampling-based pathfinding algorithm designed to efficiently discover a rapid path that is nearly optimal. RE-RRT* iteratively improves the path by utilizing the Choose Parent and Rewire processes. As the random tree expands, the sample space becomes restricted, resulting in a reduction in unnecessary searches. Simulation demonstrate that the RE-RRT^* algorithm is capable of achieving smoother and more efficient convergence to a shorter path with fewer iterations, even in the presence of various obstacles. The proposed approach can enhance search efficacy, expedite convergence, and reduce processing time. Due to these factors, our proposed RE-RRT^* algorithm has superior performance compared to previous sample-based algorithms in terms of stability, speed, computing time, and sampling space.

I. Introduction

Currently, autonomous vehicles refer to mobile robots that are utilized in many industries and for transportation purposes [1]. Developing a suitable strategy to provide an unbiased condition with little impact for robot is a crucial topic of discussion [1]–[3]. A previous study [4] has identified geometric search techniques and graph search methods. The path-planning strategies must be enhanced to possess greater intelligence and adaptability in intricate scenarios [5]. When doing forward searching in continuous coordinates, it is necessary to utilize the Rapidly Exploring Random Tree (RRT) algorithm [6]. Although this strategy is efficient in terms of speed, it lacks utility in constrained environments with complex situations. The method can be employed, based on certain decision criteria, to locate the shortest obstacle-free route [7]. For instance, the utilization of the Jump Point Search approach can significantly enhance the efficiency of the algorithm, resulting in a tenfold improvement [8]. Liu and his colleagues [9] subsequently expanded the Jump Point Search technique to a three-dimensional context. By including dynamic limits, the Hybrid algorithm [10] is able to generate pathways that are both smooth and meet the requirements of the robots. Autonomous cars were designed to be navigated in two dimensions using the algorithm [11]. Dolgov et al. [12] devised a highly integrated search technique. Their method achieves a local best solution by using local planning that involves nonlinear optimization and considers the 3-dimensional kinematics state space of the vehicle. In a previous study, Suresh and his colleagues [13] utilized FSVM to establish a safe path that avoids dynamic impediments and prevents accidents. The findings indicate that this method is effective. Chu and colleagues [14] have implemented an algorithm for real-time route planning. The trajectory generated by the Rapidly Exploring Random Tree (RRT) algorithm is expected to be inferior because of its arbitrary nature [15]. Rapidly Exploring Random Tree Star, often known as RRT*, is considered a significant advancement compared to RRT [16]. Moreau et al. [17] have devised enhanced techniques for generating curves only in scenarios like intersections. RE-RRT^* is providing a more generalized path planning solution that efficiently handles a variety of environments, not just intersections. RRT^* progressively enhances the initial path by continuously sampling. RRT^* integrates the procedures of searching for adjacent nodes and reconfiguring the tree to identify the most effective path. The RRT^* algorithm necessitates a substantial allocation of memory and time in order to determine the most optimum route [18]. To achieve the movement of two trees in the same direction, as demonstrated in the instance of RRT Connect [19], the algorithm generates two distinct trees. A satisfactory first response can be promptly generated within [20] by utilizing a 2D Gaussian mixture model. Batch Informed Trees [21] limits the range of possible states to a subset that expands gradually, allowing for quick discovery of a viable path. Nevertheless, these solutions proved to be predominantly beneficial in particular conditions. Although the APF approach is integrated into RRT* in [22] to improve the rate of convergence. Zhou W [23] proposes a new infrastructure that utilizes an improved Rapidly Exploring Random Tree (RRT) method and a linear time-varying path optimization and tracking control. The stability control of a trajectory is achieved through the utilization of the LTV Model Predictive Control (LTV-MPC) technique. The study investigates the influence of vehicle speed, planning step, and cycle on the real-time planning and stability monitoring, as stated in reference [24]. The Dubins-RRT [25] algorithm improves path smoothness and obstacle avoidance, while improved control techniques and virtual target guiding guarantee finite-time convergence of tracking errors. Evidence has shown that curve interpolation is a highly effective technique for generating reference pathways. According to Islam and Park [26], they define the method which significantly improves the accuracy of depth maps, crucial for navigating complex environments and avoiding obstacles. To further enhance the perception and decision-making capabilities of autonomous robot, the approach described by Lee, Naguib, and Islam [27], which focuses on 3D deep object recognition and semantic understanding.