Moving Robot Obstacle Recognition With Server-Based Fleet Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Moving robots face challenges in accurately recognizing new obstacles using pre-stored algorithms, leading to incorrect navigation and user inconvenience, and existing solutions require redundant learning across multiple robots and burdensome data transmission.
Innovation Solution
A moving robot generates field data and transmits it to a server for learning, where the server updates recognition algorithms based on data from multiple robots, allowing selective and efficient data transmission and learning, with the robot comparing recognition results to re-recognition results to determine which data is significant for learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a moving robot uses a pre-stored algorithm to recognize obstacles, then the robot can perform autonomous navigation, but the robot cannot accurately recognize new types of obstacles
Solution Approach 1:
The patent implements dynamic obstacle recognition by allowing the robot to adapt its recognition algorithm through continuous learning. The robot collects obstacle data during navigation, transmits it to a server for processing, and updates its recognition model iteratively, transforming a static pre-stored algorithm into a dynamic adaptive system that improves accuracy over time
Solution Approach 2:
The patent introduces a server as an intermediary between the robot and the obstacle recognition process. The server receives obstacle data from the robot, processes and analyzes this data using advanced algorithms, and generates updated recognition models. This intermediary enables the robot to benefit from centralized processing power and collective learning across multiple robots
2Adaptability or versatility
If all moving robots perform independent learning, then each robot can adapt to its environment, but redundant learning occurs across multiple robots
Solution Approach 1:
The patent merges the learning processes of multiple robots by collecting obstacle data from all robots and processing it centrally on a server. This combined learning approach allows all robots to benefit from the collective experiences of the entire fleet, eliminating redundant learning and reducing the time each individual robot would need to spend learning independently
Solution Approach 2:
The patent creates a universal learning system where a single server performs the learning function for multiple robots. The server processes data from any robot and generates universal recognition models that can be applied across the entire robot fleet, making the learning system multi-functional and efficient
3Measurement precision
If field data from all moving robots is transmitted to a server, then comprehensive learning can be performed, but data transmission burden increases
Solution Approach 1:
The patent extracts and transmits only the essential obstacle data and recognition results to the server, rather than transmitting all raw field data. By filtering and selecting only the critical information needed for learning, the system maintains high learning accuracy while significantly reducing data transmission volume and energy consumption
Solution Approach 2:
The patent implements partial data transmission by sending only the necessary subset of field data to the server. Instead of transmitting complete and excessive data sets, the system transmits just enough information to achieve effective learning, optimizing the balance between learning quality and transmission cost
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
Disclosed is a control method of a moving robot, the method comprising: a travel control operation in which the moving robot generates field data by sensing information regarding an environment around the moving robot, and controls autonomous traveling based on a recognition result that is obtained by inputting the field data to a recognition algorithm stored in the moving robot; a server learning operation in which the moving robot transmits at least some of the field data to a server and the server generates update information based on the field data transmitted by the moving robot; and an update operation in which the server transmits the update information to the moving robot.