Self-driving cars and mobile robots utilize visual sensors including cameras and Lidars to perceive the environment. Same data is utilized to generate a 3D map for localization, path planning and obstacle avoidance purposes. These vehicles are also expected to modify their path plans rapidly when an emergency vehicle such as a fire truck or an ambulance is approaching. Required steps include an early detection, determining and maneuvering to a safe spot to park. However, these steps are challenging with a visual perception only system since it requires a direct view without an obstacle in between. At this research effort, we formalized a set of experiments to analyze the robustness of location and velocity vector prediction quality of a robot equipped with a transducer array. Detection and mapping algorithms work simultaneously to mark safe and unsafe routes. Case studies include single and two robots settings with stationary, constant velocity and acceleration motion profiles while occluding planes are placed to simulate a direct and T section occlusions. Accuracy for both systems are compared with plots. It is shown that direction estimation is sufficient enough to modify the occupancy grid rapidly prior to emergency vehicle reaches to a close proximity.
展开▼
