Static warning fields used in today’s industrial Automated GuidedVehicles (AGVs) often lead to unnecessary slowdowns and stops, especially in complex environments. This thesis investigates the design of an adaptive warning field system based on predicted trajectories of both the AGV and surrounding obstacles, using 3D vision as itsmain sensing method. A modular system was developed that uses DBSCAN clustering and Kalman filtering to track obstacles and predict motion. Combined with trajectory-based collision detection, the system is used to control the AGV’s velocity. The system was tested offline using real-world data collected from a physical AGV on a test track, as well as with dynamic obstacles using synthetic point cloud data. Results showed that the adaptive system improved efficiency while maintaining safety, reduced false positives, and managed challenging scenarios, such as tight corners, more effectively. These findings suggest that predictive, vision-based safety systems may be a valuable step toward more intelligent and efficient AGV operations.