Enhancing Safety Systems for Autonomous Vehicles

2019-10-29

In order to enhance the safety of autonomous driving systems, engineers at the Massachusetts Institute of Technology (MIT) have developed a system that can perceive subtle changes in ground shadows. This system may one day be used by autonomous vehicles to quickly avoid potential collisions with other vehicles or pedestrians. Additionally, this system can also be applied to robots to prevent them from colliding with people while transporting goods. In a paper to be presented at the upcoming International Conference on Intelligent Robots and Systems (IROS), researchers describe successful experiments of autonomous vehicles driving near parking lots. When a vehicle approaches and parks, this vehicle-based system is half a second faster than traditional laser radar systems. Although this may not seem like much, researchers say it is crucial for fast-moving autonomous vehicles. "Our method can provide early warning to the AI processing system that someone is turning, allowing the vehicle to decelerate, adjust its path, and prepare to avoid a collision," said Andrew and Ena Vettori, two professors at MIT. "The ultimate goal is to provide various 'X-ray vision' capabilities for vehicles traveling at high speeds on the streets." Currently, this system has only been tested in indoor environments. Robots move at much slower speeds indoors, and lighting conditions are more consistent, making it easier for the system to perceive and analyze shadows. In their work, researchers built a system called "shadow camera" that uses computer vision technology to detect and classify changes in ground shadows. In terms of input, video frames from the camera are used to target specific areas, such as the floor in front of a corner. It can detect changes in light intensity over time, some of which may be difficult or impossible to detect with the naked eye. The system can then classify each image as containing a stationary object or a dynamic, moving object, and react accordingly if it is a dynamic image. In one test, researchers evaluated the system's performance in classifying moving or stationary targets. They used an automated wheelchair to move towards corners of a corridor to test the system's ability to recognize motion and stillness, achieving an accuracy rate of 70%. In another test, researchers installed the shadow camera on an autonomous vehicle parked in a parking lot and turned off the parking lot lights to simulate nighttime driving conditions. They compared the detection time of the shadow camera system with that of a laser radar system. In a scenario where the detection time of a car making a U-turn was measured, the shadow camera system was approximately 0.72 seconds faster than the laser radar. Furthermore, due to the researchers' adjustment of the shadow camera specifically for the lighting conditions of the garage, the system achieved a classification accuracy rate of around 86%. The researchers stated that they will further develop this system to work under different indoor and outdoor lighting conditions. In the future, there may also be methods to accelerate the shadow detection process and automate the annotation of target areas.

RECENT AI NEWS

RECENT AI TOOLS