An Approach to Real-Time Collision Avoidance for Autonomous Vehicles Using LiDAR Point Clouds

This paper proposes a novel approach for solving the problem of collision avoidance for autonomous vehicles starting from data provided by LiDAR sensors. Rather than attempting the actual recognition of pedestrians or other moving or static objects – as in the solutions based on machine learning - we define “safety bubbles” around the vehicle and all the other moving entities identified within the on-vehicle LiDAR sensing area, and issue a signal for the upper control layers when the boundary of the vehicle’s safety bubble intersects with other objects’ bubbles. The shape and size of these safety bubbles are dynamically adjusted depending on the speed of the objects. This solution is an extension/adaptation of an idea successfully applied in one of our previous works in the context of the problem of obstacle avoidance for mobile robots. The proposed algorithm was tested using CARLA simulator with promising results, as it reduces the required computational load, so that it can be used in real time, with commercially available LiDAR sensors.