Robots have been increasingly applied in orchard management, with fruit tree localization and mapping being crucial for precision and automation. Existing methods require real-time kinematic (RTK) global navigation satellite system (GNSS) receivers. However, signal strength is diminished in orchards due to being occluded by tree canopies. Therefore, this study proposes a novel system for global fruit tree positioning and mapping in semi-structured orchards based on multi-sensor fusion simultaneous localization and mapping (SLAM), without dependence of positioning signals from satellites. As the specific contributions, firstly, a SLAM system which fuse a camera, an inertial measurement unit (IMU), and a light detection and ranging (LiDAR) sensorwas constructed to improve the accuracy of robot odometry estimation in complex orchard environments. Secondly, a fruit tree localization algorithm was developed to localize the fruit trees around the robot using both images and LiDAR point clouds, after which the global positions of the detected fruit trees were optimized using the SLAM-derived robot pose real-time. Our method has been extensively evaluated in real pear and persimmon orchards. The results showed that the average root mean square error (RMSE) for odometry was 6.04 cm and 14.59 cm in pear and persimmon orchards, respectively, and the positioning errors of the detected fruit trees were 0.57 cm and 3.23 cm, respectively. The results indicate that our method can achieve accurate positioning and mapping of fruit trees in complex environments and build fruit-tree maps.
2 December 2024 in Computers and Electronics in Agriculture
Path planning is a critical process in mobile robot navigation. Sampling-based path planning algorithms represented by Rapidly Exploring Random Tree star (RRT*) have gained widespread adoption due to their asymptotic optimality and proven efficiency. However, when applied to intricate environments, characterized by narrow passages and cluttered obstacles, these algorithms encounter challenges in both the initial solution generation and the convergence towards the optimal path, mainly caused by the inefficient sampling strategy, thereby impeding its overall effectiveness. To address these limitations, we introduced Agile-RRT* (A-RRT*), an advancement of RRT* algorithm. Our contributions are twofold: firstly, we introduce an adaptive goal-biased sampling strategy, which employs an adaptive principle for determining the step size on the basis of the goal-biased strategy. This avoids getting trapped in local minima and enhances the efficiency of the initial solution generation. Secondly, we introduce a path optimization approach using a secondary tree and subset-informed sampling, to accelerate the convergence toward the optimal path. It optimizes the path by gradually shrinking the designed elliptical planning space around local states, which effectively narrows down the search space. The experimental results demonstrated that the proposed A-RRT* diminishes the initial solution search time by 71.00% and the sub-optimal solution search time by 82.86% in comparison to RRT*. The A-RRT* exhibits superior performance over RRT*, Informed-RRT*, P-RRT* and Quick-RRT* in terms of soundness and efficiency in narrow and intricate environments. This method could expedite efficient motion planning for drones and mobile robots in complex environments.
Typical SLAM systems adhere to the assumption of environment rigidity, which limits their functionality when deployed in the dynamic indoor environments commonly encountered by household robots. Prevailing methods address this issue by employing semantic information for the identification and processing of dynamic objects in scenes. However, extracting reliable semantic information remains challenging due to the presence of motion blur. In this paper, a novel visual SLAM algorithm is proposed in which various approaches are integrated to obtain more reliable semantic information, consequently reducing the impact of motion blur on visual SLAM systems. Specifically, to accurately distinguish moving objects and static objects, we introduce a missed segmentation compensation mechanism into our SLAM system for predicting and restoring semantic information, and depth and semantic information is then leveraged to generate masks of dynamic objects. Additionally, to refine keypoint filtering, a probability-based algorithm for dynamic feature detection and elimination is incorporated into our SLAM system. Evaluation experiments using the TUM and Bonn RGB-D datasets demonstrated that our SLAM system achieves lower absolute trajectory error (ATE) than existing systems in different dynamic indoor environments, particularly those with large view angle variations. Our system can be applied to enhance the autonomous navigation and scene understanding capabilities of domestic robots.