In modern industrial environments, robots must possess adaptive capabilities to achieve efficient movement and obstacle avoidance. Locomotion is central to robotic applications, and effective obstacle avoidance is crucial. This paper addresses the dynamic obstacle avoidance challenges faced by industrial robots by proposing a local avoidance algorithm based on the direction of obstacles and robot movement. The algorithm sets a reaction time for the robot, calculates the perception range, and within this range, predicts the likelihood of collisions based on the movement states of obstacles and the robot. Different avoidance strategies are applied according to the collision predictions, with real-time adjustments made as motion states change. By predicting collisions in advance and minimizing the use of full stop strategies, the proposed algorithm in this study improves the obstacle avoidance efficiency and reduces safety issues.