AGV is widely used in warehousing and flexible production lines. In order to make AGV operate more safely in complex environments, obstacle avoidance function is essential.
 
Obstacle avoidance refers to the process in which an AGV senses static or dynamic obstacles along its path through sensors during walking, achieving emergency braking or avoiding obstacles.
 
Today we will introduce visual obstacle avoidance。
Vision sensor
 
There are many commonly used computer vision solutions, such as depth cameras based on ToF, depth cameras based on mechanism light RGB binocular vision, etc. The working principle of ToF depth cameras is basically similar to that of laser sensors (as shown in Figure 5), which measure distance by calculating the flight time of the laser. The difference is that laser sensors scan point by point, while ToF cameras obtain depth information of the entire image simultaneously, Structured light is a method of projecting light with structural features onto the object being measured through an infrared laser. Due to different distances, the position captured by the camera is also different. By calculating the offset of the light spot in the captured image and the calibrated standard pattern at different positions, combined with parameters such as camera position and sensor size, the distance between the object and the camera can be obtained. Both ToF and structured light belong to active ranging. Binocular vision belongs to passive ranging, and its essence is also the triangulation ranging method, which obtains depth information through the disparity between two cameras. The advantages of visual sensors are their wide detection range, rich information acquisition, and lower price compared to laser sensors. The disadvantage is that they require high computing power, ToF is affected by multiple reflections, structured light is affected by reflections, and binocular vision is affected by changes in tube lighting and physical textures. Each has its own advantages and disadvantages. In practical applications, a variety of different sensors are generally integrated to ensure that AGV robots can correctly perceive obstacle information in different applications and environmental conditions. With the increasing intelligence of AGV robots, the requirement for obstacle avoidance is no longer as simple as parking. After sensors obtain obstacle information, they can use obstacle avoidance algorithms to update the target trajectory in real time and bypass obstacles. Traditional obstacle avoidance algorithms include bug algorithms, potential field methods (PFM), vector field histograms (VFH), and other intelligent obstacle avoidance technologies such as neural networks and fuzzy logic.
 
Any interest, please feel free to contact us for more details.
Email: inquiry@dflyaitech.com