Sewing Welding Positioner kuka Robot Sushi Desktop Robot Sushi Welding Positioner Rotation range: ±360
Robot positioning sensor
Robots also require sensors for positioning, which are mainly divided into two categories: LiDAR and Vision
In the early years, LiDAR was introduced by Western companies Velodyne, Hokuvo, and SICK. However, with the rapid development of science and technology in China this year, exciting LiDAR products have emerged. Navigation and positioning also require other sensors, such as global positioning, radar sonar, and visual depth camera sensors
Different sensors have different advantages and disadvantages. Let me talk about the comparison between LiDAR and vision. The point cloud generated by LiDAR needs to be matched or correlated with several point clouds in the map for global positioning. The advantages are high accuracy and three-dimensional information; The disadvantage is the lack of texture information. In contrast, vision provides texture information that LiDAR cannot provide, but its accuracy is lower. It can achieve positioning accuracy of tens of meters or tens of meters outdoors, while it can reach centimeter level indoors. Of course, one of the biggest problems with vision is the loss of three-dimensional information.
I just talked about what global positioning is and what is the difference between it and local positioning; At the same time, we also talked about sensing technology. Next, we will talk about global positioning technology, which is the algorithm used to complete this work.
The problem of global positioning has a research history of more than 20 years. It is interesting that the initial global positioning method was started with LiDAR, mainly using traditional manual features, that is, processing and extracting features from LiDAR point clouds, and completing global positioning through feature matching. The matching algorithm is based on optimization methods. At present, the development direction is mainly based on vision to achieve global positioning. The method of feature extraction mainly relies on deep learning, and it utilizes the extraction and matching of local features to achieve global positioning
In terms of method classification, global positioning technology has three dimensions:
1. Sources of sensing information: LiDAR, vision, and multimodal fusion.
2. Feature extraction method: Traditional manual model 1 requires feature extraction and deep learning due to the large amount of raw data and often poor discrimination.
3. The localization algorithm uses the extracted features to complete the localization: optimization method - when the localization loss function reaches the minimum value, it completes the localization for the robot's position, and completes the localization through panoramic descriptor matching or local feature matching.
About the company
