Abstract: Although multiple thread colors and stitch patterns can be precisely controlled by fully autonomous embroidery, the low quality of edge contour extraction and the unreasonable embroidery path design lead to a sharp increase in thread breakage rate, seriously affecting the quality of finished products. Therefore, an autonomous planning framework for complex embroidery process constraints is proposed to alleviate the lack of flexibility in small-batch embroidery. Firstly, an improved Canny operator is used to extract edge contour information. Combined with the inherent curvature constraints of the embroidery machine and the continuous stitch trajectory constraints, a region-wise embroidery path optimization model without thread breakage is constructed. Furthermore, a curvature/tension-driven destroy-and-repair operator is introduced, and an adaptive large neighborhood search algorithm is designed to ensure global embroidery quality. Finally, experimental tests on various patterns, including animals and text, show that the proposed method significantly improves the embroidery efficiency and stability.