Camera mounted on the robot continuously take photos of the environment. From these photos we can extract recognizable and unique features. Base on the number of features in image frames and the similarity between frames, we can pick out the most representative data, namely the key frames, from large amounts of image information, and transmit these key frames, vehicle posture information and matching information between key frames to the cloud server via internet.
After the cloud server obtained key frame information, vehicle posture information and matching information between the key frames,with diagram optimization method it establish the equation to all the databased on the observation model and the kinematic model, furthermore,it uses iterative method to minimize the error to build an off-line 3D map with visual features,
In the process of navigation, robots continuously transmit key frame information to the cloud server. The cloud server then base on the matched degree among key frames, feature appearing frequency and the weight of dynamic adjustment features, discard the features with low observation frequency and continuously optimizes the 3D position with high observation frequency features.
For forklift type robot scheduling is very difficult. Different from other vehicle-scheduling problem, forklift robot scheduling problems are as follows:
① Travel distance is relatively short, which is different from car scheduling because car can be regarded as a point on the road.
② Collision, livelock, and deadlock issue happen easily due to the narrow space and low road capacity.
That is the reason why the forklift robot cannot use normal car scheduling method. Instead, the forklift robot must use customized scheduling methods with a large number of calculations. Thus, the cloud server is indispensable. We use the time window method and area control method to schedule the forklift robot from the cloud.
1.Cost-effective.
2.Easy to use.
Before leaving the factory, robots will go through a lot of off-line tests and obtain rough operational parameters by systematical identification method. In actual operation, control system will use predictive control and self-adaptive control methods to tweak and optimize parameters on-line based on operation parameter obtained off-line, and adjust controlling demands dynamically, so as to control quickly and precisely.
Most of the forklift robot manufacturers on the market normally identify their product performance of with stopping accuracy. From their point of view, the only way to fork pallets correctly is to raise robot’s stopping accuracy and limit cargo’s placing location, which means to have extra limit mechanism installed to ensure high precision forking accuracy. However, our vision guided forklift don’t follow those ideas. When forking the goods, 2D laser or 3D TOF camera will observe and identify the pallet, and generate the optimal path based on real-time feedback data to self-adapt itself to fork the goods.
1.Quick implementation without extra limiting mechanism.
2.Easy to operate, user-friendly human-machine collaboration.
3.Stable and efficient for smooth operations.
Camera that robot carries can provide continuous environment image so that we can extract recognizable and unique information from fixed objects such as walls and shelves, artificial labels such as QR codes, and geometries such as lines and cross points. Such information become unique tags of specific objects in external environment, just like the road signs and doorplates in the city of the real world, enabling robots to judge their own location.
After matching the off-line map with onsite installation layout and marking specific points, robots can calculate the optimal path to go to the destination in a map based on current running condition of multiple vehicles controlled by the system.