We can oftensee this scene:from the first golden ray of the morning sun, to the last deep blue of the night, unmanned cleaning robots silently travelling on the road, guarding the cleanliness of the city.
Among the many elements that support the robot's ability to work around the clock, perception technology is a key part of it. If only one perception method is used, there will be some limitations: pure visual perception in the light changes and bad weather recognition ability is limited, it is difficult to properly deal with the corner cases (edge cases); pure radar perception although all-weather advantages, but the point cloud data confidence and semantic information is insufficient, easy to affect the accuracy of target recognition ......For this reason, the unmanned cleaning robot Star Hyun® innovatively adoptsa multi-source heterogeneous sensor solution- vision AI-led, combined with a variety of sensors working in concert, and advanced perception algorithms for processing and fusion of data, to effectively improve the robot's ability to adapt to the environment and the accuracy of target recognition.This solution not only enhances the robot's long-term operational stability and intelligent decision-making ability, but also ensures that the robot can efficiently and safely perform cleaning tasks in diverse urban environments through continuous learning and optimisation.Holographic Insight: Multi-Source Heterogeneous Sensor Solution
StarKeun®'sMulti-Source Heterogeneous Sensor Solutionintegratesnon-contact sensorsandcontact sensorsto achieve omni-directional and multi-layered environmental sensing.
The non-contact sensors include high-definition visible cameras, infrared thermal imaging cameras and LiDAR, forming a "multi-dimensional sensing data matrix".◉ High Definition Camera:With high resolution and wide viewing angle, it is capable of capturing clear RGB images, presenting every detail of the surrounding environment, including colours, textures and dynamics, providing rich visual information for Star Kyunney®;◉ Infrared Camera:At night or in low light conditions, infrared thermal imaging technology is able to penetrate the darkness, detect differences in heat, and identify hidden obstacles or areas of abnormal temperatures, ensuring that the system maintains efficient sensing capabilities in all lighting conditions;◉ LIDAR:It can provide high-precision 3D point cloud data for accurate ranging and obstacle detection. Whether indoors or outdoors, LIDAR can be unaffected by light conditions and provide stable and reliable sensing performance.Contact sensors, on the other hand, acquire hardness data through force contact, which not only protects the operating mechanism, but alsoprovideseffective supervisory informationforvisual perception.The data from these sensors are not isolated, but fused by algorithms to form a unified, high-dimensional perception picture, so that each pixel has information such as colour, texture, hardness, etc.The fusion algorithmsare thenused to process the heterogeneous data from multiple sources, which can improve the precision and robustness of perception, and significantly enhance the robot's ability to adapt to complex environments.Fusion perception: multimodal data fusion algorithm
Fusion algorithms play a critical role in the autonomous driving system of the unmanned cleaning robot Star Kyunney®, which is responsible for processing and interpreting the massive amount of information from the "multi-dimensional heterogeneous data matrix" to provide accurate environmental understanding and prediction.◉ Adaptive environmental state: Through the cross-checking of feature vectors of data from multiple sensors, such as far infrared, visible light, LiDAR, etc., it can adaptively cope with different scenarios, such as backlighting, nighttime, rain and fog;
◉ Target recognition and classification:identifying pedestrians, vehicles, animals, static obstacles and traffic signs in the operation scene, accurately identifying traffic signals such as traffic lights, and providing a decision-making basis for the robot's actions;
◉ Sc ene creation and semantic segmentation:create a detailed scene map using point cloud data and image information collected by sensors, semantically segment it, and identify key elements such as drivable areas, pavements, traffic signals, etc;
◉ Dynamic target prediction:in the face of complex scenes such as people or other objects intruding into the scene, the algorithm outputs the current state in real time and predicts the future actions of the surrounding objects based on the historical behavioural patterns and the current environmental conditions, which makes it easier for the robot to make more appropriate emergency response and operation strategies;
◉ Unknown small object analysis:through the continuous multi-image synthesis of various bump changes on the road in front of the vehicle, so as to accurately perceive the scale of unknown obstacles, and judge the hardness of the object through the contact sensors, anddecide whether to "pass" or "put away". The hardness of the object is determined by a contact sensor, which determines whether to "pass" or "retract" the sweeping mechanism.
In the development of unmanned cleaning robots, the mutual matching and promotion of sensor technology and perception algorithms is the core of technological innovation, and its application not only improves the operational efficiency of the robot, but also provides a solid guarantee for the safety and reliability of the robot.
