By integrating vision cameras, LiDAR and infrared cameras, the robot is able to perceive its surroundings in different dimensions, enabling it to manage the quality of its work in a refined manner, while a real-time feedback system ensures that the robot is able to respond flexibly to changes in the dynamic environment.

Distribution of body cameras

To ensure the accuracy of the data, the data collected by the sensors are pre-processed, such as de-noising, filtering and normalisation, to ensure the accuracy and consistency of the information; the processed data are then feature extracted by neural networks, allowing the robot to recognise environmental features and analyse the current state.

Based on a large number of annotated datasets from reality and generative AI simulations, the deep learning model will conduct extensive training on force control strategies for different environments to improve the robot's generalised reasoning ability, so as to better cope with all kinds of complexities encountered in operations.

In actual operation, real-time sensor data will be input into the trained model, which will instantly predict the appropriate force control strategy according to the current environment state and precisely execute the corresponding action through the controller. At the same time, the force sensors continuously monitor the force exerted by the robot during operation, transmit the feedback data to the control system, and dynamically optimise the control strategy according to the actual situation to ensure that the robot is equipped with an adaptive and independent sweeping disc that can be executed in place.

Just as a human being adjusts the force when touching an object, the tactile sensors enable the robot to be sensitive to changes in the environment and precisely adjust its sweeping action. For example, when sweeping close to edges, the robot automatically adjusts its force to avoid excessive disc intrusion, thus reducing wear and prolonging service life.

Through continuous monitoring and dynamic feedback, the robot forms a closed loop of perception-decision-execution, ensuring efficient and reliable operation in complex environments.

Reinforcement learning plays a crucial role in this process. Through continuous interaction with the environment, the robot is able to learn the optimal strategy for different scenarios. For example, during the cleaning process, the robot accumulates operational data, identifies the pattern of rubbish distribution, and gradually optimises the cleaning path to improve cleaning efficiency; the intelligent cloud control platform further enhances the robot's self-optimisation capability.

Robot self-optimisation process

In actual operation, the robot uploads real-time sensory data to the intelligent cloud control platform; the platform analyses a large amount of historical data to provide the robot with optimisation suggestions and update the cleaning strategy; the robot then adjusts and iteratively optimises its own decision-making mode based on the feedback from the platform.

This combination of self-adaptive and reinforcement learning capabilities makes the robot no longer a simple automation tool, but an intelligent body with the ability to evolve itself.

Through data analysis and real-time monitoring, the robot can ensure that the quality of work strictly follows the sanitation assessment standards, and guarantee the continuity and consistency of work through real-time monitoring. In addition, the intelligent cloud control platform provides detailed feedback and assessment of the quality of the work, improving decision-making efficiency and achieving perfect integration with the requirements of the cleaning operation.

Robot in the process of operation

With the continuous development of technology, unmanned cleaning robots are gradually getting rid of the limitations of traditional automation and moving towards the era of true intelligence. In the future, KUSA Technology will continue to deepen the embodied intelligence technology, not only in the field of unmanned cleaning, but also in more complex scenarios to achieve the depth of interaction between the robot and the environment, to provide more intelligent solutions for urban services.