By detecting subtle changes in the reflection, the new system is able to detect incoming vehicles, pedestrians or other obstacles that may cause a collision at the corner.
Editor’s note: This article comes from the WeChat public account “MEMS” (ID: MEMSensor), author Memes Consulting Yin Fei, the original title “Farewell to the ghost probe”, MIT developed a perception system that can “see” the corner behind the corner “, slightly cut.
In order to further improve the safety of the autopilot system, researchers at the Massachusetts Institute of Technology (MIT) developed a method to sense subtle changes in ground reflections to determine if there are moving objects at the corners, according to Memes Consulting. New system.
In the future, autonomous vehicles can use this system to avoid potential collisions with another car or pedestrian at the corner where the line of sight is blocked. In addition to self-driving car applications, drugs or item transport robots that navigate in the hospital corridor in the future can also use the system to avoid hitting pedestrians.
In a recent paper presented at the International Conference on Intelligent Robotics and Systems (IROS), MIT researchers introduced successful experiments using autopilot cars in garages and automatic navigation wheelchairs in corridors. This new system successfully defeated the traditional Lidar (LiDAR) system for the perception of the car at the corner, because the latter can only detect objects in the “field of view”. Compared to lidar, MIT’s new system has been perceived more than 0.5 seconds earlier.
The researchers said that this does not seem to be much, but for autonomous cars driving at high speeds, 0.5 seconds may mean whether the collision can be avoided.
“For robotic applications that operate in environments where there are other moving objects or people around, use our system to alert the robot in advance to alert pedestrians to approach, thus controlling the robot to slow down, adjust the path and avoid it in advance “The preparation for the collision,” added Daniela Rus, co-author of the paper, director of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and professor of electrical engineering and computer science, Andrew and Erna Viterbi. “Our ultimate goal is to drive fast on the road. Vehicles provide this X-ray-like ‘perspective’ capability.”
Currently, the system has only been tested in an indoor environment. Indoors, robots move much less quickly and lighting conditions are simpler, which helps the system to detect and analyze reflections more easily.
The system uses a standard optical camera that uses a series of computer vision techniques to monitor changes in reflected light intensity to ultimately determine whether the reflection is projected by a moving or stationary object and predicts the possible path of movement of the object in question.
In a separate test, the researchers installed the “ShadowCam” system they developed in a self-driving car in the parking lot, and the headlights in the parking lot.They are all closed to mimic the night driving environment. They compared the ShadowCam system to the lidar. In an example scenario, the ShadowCam system detects 0.72 seconds faster than a laser for a car that is turning around a column.
Image Source: MEMS
In the previous test, an auto-running wheelchair equipped with the ShadowCam system determined whether a pedestrian was approaching by detecting the reflection of a person projected on the green area at the corner.
However, the experiment has so many limitations so far: for example, the experiment was only tested under indoor lighting conditions, and the research team also needed a lot of work to adapt the system to higher speeds and Complex lighting conditions for outdoor lighting. Despite this, the system allows for a promising prospect that ultimately helps autopilots better sense pedestrians, cyclists, and other vehicles on the road.
Next, the researchers will further develop the system to work in different lighting conditions indoors and outdoors. In the future, new methods will be developed to speed up the system’s reflection detection and automate the annotation of the reflection-aware target area.
This research was funded by the Toyota Motor Research Institute.