In general, AGV products are divided into two categories, one is non-reflector laser forklift AGV, and the other is magnetic navigation AGV cart. The greatest difference between the two kinds of AGV products is the navigation mode. AGV cart adopts magnetic navigation method, which is a marked navigation technology; while the non-reflector laser forklift AGVA adopts AGV unique non-reflector laser navigation technology, which is a non-marked navigation technology.
The main components of the non-reflector laser forklift AGVA are the forklift body, non-reflector laser navigation module, safety module and vehicle control module. The forklift AGV need not lay any auxiliary mark including reflector, magnetic stripe, magnetic nail, two-dimensional code, etc.. By scanning the operating environment of the current forklift AGV with a vehicle laser scanner, the forklift AGV current position can be determined according to SLAM algorithm.
The non-reflector laser independent navigation forklift AGV can be divided into stacking type laser forklift, handling type laser forklift and counterweight type laser forklift according to the structural characteristics of the forklift body. Different specifications of forklift body can be used for each type of forklift according to the load and lifting height, etc.. If the lifting height of materials is high, generally stacking forklift AGV is used; and if the material is heavy, basically it does not need handling type forklift AGV; the difference of counterweight type forklift AGV from the aforesaid two kinds of forklift AGV is that there is no leg at the lower part of the fork, and when there is no space in the lower part of the material such as a “田” type of tray, a counterweight forklift AGV is used.
AGV cart includes motor tractor, latent cart and carrying cart. The tractor type is simple, without division; its main feature is that, when the chassis height of skip car is lower than 300 mm, the AGV cart can drive the cart to operate through the traction system, to complete the specified action; the latent models are most widely used, which can be divided into single-drive one-way latent type, single-drive two-way latent type, and dual-drive two-way latent type according to the number of drives and walking direction. The feature of this type of application is that, when the chassis height of the skip car is higher than 300 mm, a traction mechanism is mounted at the bottom of the skip car. The latent AGV cart arrives at below the skip car, and connects the traction mechanism through a body lift rod, to drive the skip car to work. According to comprehensive judgment of the operating process, site space and material load, single-drive one-way type, single-drive two-way type, or dual-drive two-way is selected. This cart can save space. The carrying cart is most complicated, since the transfer mechanism of a carrying cart can be varied, and appropriate carrying mechanism is selected according to the actual material state. Typical models include roller type, chain type, lift type and platform type. Another feature of AGV is heavy-load carrying type. Several tons of materials can be automatically handled by heavy-load carrying type AGV.
Regardless of non-reflector laser forklift AGV or AGV cart, the selection should consider the customer on-site skip car or the tray type and size, space, aisle width, material load, workflow, docking plane height, etc.. The selection of AGV should fully consider the application environment. AGV’s engineers will give recommendation of models after a comprehensive assessment of the application environments.
1. What is SLAM?
Slam (Simultaneous Localization And Mapping) is to map the robot’s own positioning (such as AGV) and surrounding environment by using sensors. When a robot comes to a strange environment, it needs to establish the relationship with the environment quickly, such as a series of questions, “Where am I now?" "Where am I going?" "Where do I come from?" "What am I going to do now?" "What is the world I see now," "Can I locate my position in the existing abstract world?" According to these questions, the robot can give perfect answers, that is, what is solved by the SLAM method.
2. Four elements for realization of SLAM method
The SLAM method should consider following four aspects when realization:
The map represents problems, which needs to be decided according to the needs of actual scenario.
Information perception, which needs to consider comprehensive external environment; usually the SLAM external environment sensors are divided into visual navigation sensors such as cameras, and laser navigation sensors, such as laser scanners.
The data association. The data types, timestamps, coordinate system expressions of different sensors are different, which need to be unified.
Positioning and composition of map, refers to how to achieve pose estimation and modeling. It involves a number of mathematical problems, physical model establishment, state estimation and optimization, etc..
3. SLAM classification
SLAM can be divided into laser SLAM (also divided into 2D and 3D) and visual SLAM.
In terms of laser SLAM 2D, that is, only single-line laser sensor is used; and on the plane of laser sensor scanning, two-dimensional positioning is performed, after acquiring precision two-dimensional positioning, three-dimensional laser point cloud is resolved, to become complete spatial 3D data.
Similarly, the laser SLAM 3D is to acquire 3-D data using three-dimensional laser sensor, and then locate through the matching of characteristic points of the 3D data, and then on the basis of three-dimensional positioning, calculate and match the complete three-dimensional data, to finally get its pose.
The visual SLAM system is divided into four modules: camera, back-end, mapping and loopback detection. Its process is as follows: The visual SLAM uses a camera to perceive the surrounding world and assess its own position. It needs to calculate the attitude parameters from adjacent two photos by the visual odometer and estimate the distance angle and the position of the points on the photo. Since there is cumulative error that needs backend optimization, the perception of the surrounding environment needs mapping and registering, and then the loopback detection and calculation of errors and correction, finally the accurate robot pose is obtained.
4. Future of SLAM
The future development trends of SLAM: one is the development towards lightweight and miniaturization. The SLAM can be operated in an embedded type or small devices like mobile phone, to complete its applications as the underlying functions. In most occasions, the real purpose is to achieve the functions of robots and AR/VR devices, such as sports, navigation, teaching, entertainment, while SLAM provides pose estimation for the upper applications. In these applications, we do not want SLAM to occupy all the computing resources, so there is an urgent need for the SLAM miniaturization and light weight. The other is to achieve precision three-dimensional reconstruction, scene understanding and other functions by using the high-performance computing equipment. Among these applications, our purpose is to perfectly reconstruct the scene, without limitation on the computing resources and equipment portability.