Abstract: In order to address the issues of crop damage and low operational efficiency caused by imperfect path planning strategies for the weeding components of small-scale intelligent weeding machines, this study utilized soil-cultured lettuce as the experimental subject. The research proposed an adaptive seedling-avoiding weeding operation path planning method based on a heuristic search strategy and concurrently designed a bidirectional parallel drive mechanism to enhance the functionality of the weeding machine. A comprehensive weeding operation platform was constructed by integrating an image acquisition system with the newly designed bidirectional parallel drive mechanism. The Denavit-Hartenberg (D-H) parameters method was employed to perform kinematic analysis of the bidirectional parallel drive mechanism, enabling precise control over the movement of the weeding components. This precise control is crucial for minimizing crop damage while ensuring effective weed removal. The study applied the YOLOv8n-seg model to accurately identify and extract the edge morphology of both lettuce and weeds from the acquired images. To protect the lettuce, a minimum enclosing circle was delineated to define the lettuce protection zone, while the positions of the weeds were represented using their centroid coordinates. This approach provided a clear and concise method for mapping weed locations within the field.Building upon the identified weed positions, an adaptive recursive clustering algorithm was utilized to partition the weed locations into distinct clusters. This clustering facilitated the construction of a weighted undirected graph for the weeding component’s operation path, with the cluster centers serving as key points within the graph. By integrating this graph-based approach with the A* (A-Star) heuristic search strategy, the study successfully achieved the shortest possible path planning for the weeding components under seedling-avoiding conditions. This integration ensures that the weeding machine can navigate efficiently around the protected lettuce areas while effectively targeting weed clusters. To validate the effectiveness and adaptability of the proposed path planning method, extensive simulation analyses were conducted under various weed distribution scenarios. Additionally, performance comparison experiments were carried out under identical weed distribution conditions to benchmark the proposed method against traditional planning algorithms. The simulation results demonstrated that the proposed path planning method consistently achieved coverage of weed growth areas with shorter paths across different weed distributions, highlighting its strong adaptability and efficiency. The operation path lengths, ranked from shortest to longest, are as follows: the proposed method, the Spiral algorithm, the Zigzag algorithm, and the BCD algorithm. Specifically, under discrete weed distribution states with grid coverage rates of 20%, 40%, and 60%, the proposed path planning method outperformed other traditional planning algorithms. Compared to the BCD algorithm, the path lengths were reduced by 54.6%, 53.3%, and 60.5% respectively. These significant reductions in path length not only enhance operational efficiency but also contribute to minimizing fuel consumption and reducing wear and tear on the weeding machinery. Experimental results further corroborated the simulation findings. The adaptive seedling-avoiding weeding mode achieved a weeding rate of 93.91%, indicating a high level of effectiveness in weed removal. Importantly, the crop damage rate was maintained at 0, demonstrating the method’s precision and its ability to protect the cultivated lettuce from unintended harm. Additionally, the average weeding duration per unit, defined by spatially adjacent lettuce plants in both row and column directions, was measured to be 14.2 seconds, indicating a significant improvement in operational speed. When compared to the seedling-avoiding exhaustive weeding mode, the adaptive mode enhanced operational efficiency by 42.28%, showcasing its potential for practical application in agricultural settings. The integration of advanced image processing techniques, precise kinematic control, and sophisticated path planning algorithms in this study provides a robust framework for improving the performance of small-scale intelligent weeding machines. The proposed adaptive seedling-avoiding weeding operation path planning method not only addresses the critical issues of crop damage and low efficiency but also lays a foundation for future advancements in mechanical weeding technologies. In conclusion, the research findings offer a valuable technical reference for achieving high-quality mechanical weeding in field applications. By combining heuristic search strategies with adaptive clustering and precise motion control, the study presents a comprehensive solution that enhances both the effectiveness and efficiency of weeding operations. These advancements hold significant promise for the agricultural industry, contributing to more sustainable and productive farming practices through the adoption of intelligent weeding machinery.