Obstacle recognition information feedback method and apparatus, robot, and storage medium
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ground sweeping robots face challenges in accurate obstacle recognition due to limited viewing angles and reliance on specific picture libraries, leading to instances of inaccurate or missed recognition.
Innovation Solution
A method and apparatus for obstacle recognition information feedback, which allows users to submit feedback information, including pictures and type information, to a server or self-walking robot, enabling the enhancement of obstacle recognition models through user correction and de-identification of sensitive content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If obstacle recognition is trained only with pictures from the robot's viewing angle, then the system complexity is reduced, but the recognition accuracy deteriorates due to limited viewing angles
Solution Approach 1:
The patent implements a feedback mechanism where users can provide correction feedback for misrecognized obstacles. The system collects user feedback on recognition errors, uses this feedback to generate corrected training data, and retrains the recognition model iteratively. This closed-loop feedback process continuously improves recognition accuracy without requiring complex multi-angle data collection systems.
Solution Approach 2:
The system enables users to actively participate in improving the recognition model by submitting feedback on misrecognized obstacles. Users serve as human annotators who correct recognition errors, and the system automatically processes this user-generated feedback to create training data. This self-service approach allows the model to improve itself with minimal system complexity.
2Measurement precision
If a large amount of labeled data is used for training, then the intelligent determination ability is improved, but the time and resources required for data preparation increase
Solution Approach 1:
The system allows users to actively contribute labeled training data through the feedback mechanism. When users correct recognition errors, they are effectively labeling data that becomes part of the training set. This user-driven data labeling process eliminates the need for manual data preparation by developers or researchers, continuously expanding the training dataset without additional time investment from the system operators.
Solution Approach 2:
The feedback collection and model retraining process operates continuously as the system is used. Each user interaction potentially generates new training data, and the model can be retrained incrementally. This continuous process ensures the model improves over time without requiring periodic large-scale data preparation campaigns, maintaining both accuracy and efficiency.
3Measurement precision
If user feedback is collected to improve recognition, then the recognition accuracy is improved, but the system complexity increases due to feedback management
Solution Approach 1:
The feedback collection interface is designed to be simple and user-friendly, requiring minimal effort from users. The system automatically processes the feedback data, performs data cleaning, annotation, and model retraining without requiring complex feedback management infrastructure. Users simply provide corrections, and the system handles all the complexity of managing this feedback internally.
Solution Approach 2:
The patent introduces a server as an intermediary that manages the complex feedback processing tasks. The server receives feedback from multiple users, aggregates and cleans the data, performs model retraining, and updates the recognition system. This intermediary architecture centralizes the complexity of feedback management, keeping the user interface simple while handling sophisticated data processing in the background.
Data Source
AI summary
Provided are an obstacle detection method and apparatus, a self-walking robot, and a storage medium. The obstacle detection method includes: upon reception of a triggered feedback instruction, providing an interactive interface to allow a user to submit feedback information, the feedback information including a related picture and type information of an obstacle contained in the picture; and after learning that submission by the user is completed, sending the feedback information to a server or notifying a bond self-walking robot to send the feedback information to the server.