To develop and validate a 3D U-Net algorithm for automated pedicle screw planning in thoracolumbosacral regions.The model was trained on 1,235 retrospective cases (1,005 public datasets [CTSpine1K] and 230 clinical cases from Beijing Tongren Hospital), including 1,165 for training (160 clinical + public data) and 70 for validation. Performance was assessed using Dice coefficient for spinal segmentation, Gertzbein-Robbins scale for screw accuracy, Babu scale for facet joint invasion, and Kappa statistics for inter-observer consistency.In 70 patients, 840 T12-S1 screws were automatically planned. Segmentation achieved a Dice coefficient of 0.9495. Screw accuracy showed 98.8% Gertzbein-Robbins grade A (830) and 1.2% grade B (10), with no C-E grades. Facet joint invasion grades were 96.43% grade 0 (810), 2.38% grade 1 (20), 0.95% grade 2 (8), and 0.24% grade 3 (2). Kappa values indicated strong agreement for Gertzbein-Robbins grading (κ=0.661, p<0.001) and facet assessment (κ=0.878, p<0.001). Algorithm runtime averaged 26 seconds for T12-S1 segmentation and 2 seconds per screw plan.The 3D U-Net algorithm enables rapid, accurate automated pedicle screw planning with high clinical feasibility. It demonstrates robust segmentation precision (Dice >0.94), excellent screw placement accuracy (98.8% grade A), and efficient processing times, supporting its potential to optimize robot-assisted spinal surgery workflows.Copyright © 2025. Published by Elsevier Inc.