Adaptive Robot Object Detection for Changing Illumination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods for robots face challenges in maintaining high identification accuracy when conditions such as illumination change or when target objects are piled, as they require frequent tuning to adapt to new environments.
Innovation Solution
An object detecting method that involves imaging target objects, recognizing their position and posture, counting successful recognitions, calculating a task evaluation value, updating an estimation model based on this data, and adjusting the imaging position and posture of the camera to improve recognition accuracy and adapt to changing conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the robot uses a fixed object detection method, then the system complexity is low, but the recognition accuracy deteriorates when illumination or object state changes
Solution Approach 1:
The patent implements dynamic adaptation by continuously updating the object detection model based on recognition results and task outcomes. The system transitions from a fixed detection method to a dynamic one that adjusts imaging parameters and model characteristics in response to changing conditions such as illumination and object state, thereby maintaining high recognition accuracy without requiring complex manual recalibration procedures
Solution Approach 2:
The system performs self-updating of the detection model using its own recognition results and task evaluation data. By automatically adjusting the model based on accumulated experience from successful and unsuccessful tasks, the system improves its own performance without external intervention, resolving the contradiction between maintaining accuracy and avoiding complex manual tuning procedures
2Measurement precision
If the robot frequently recalibrates the detection system, then the recognition accuracy is maintained, but the productivity decreases due to time loss
Solution Approach 1:
The patent implements continuous model updating based on each recognition result and task outcome. Instead of periodic recalibration that stops production, the system continuously learns from each interaction, accumulating experience without interrupting the workflow. This continuous improvement approach maintains accuracy while preserving productivity by eliminating downtime associated with frequent manual recalibration
Solution Approach 2:
The system establishes a feedback loop where recognition results and task outcomes (success/failure) are fed back to update the detection model. This automatic feedback mechanism allows the system to adapt to changing conditions in real-time without requiring manual recalibration, thereby maintaining high recognition accuracy while avoiding productivity loss from frequent system stops
3Measurement precision
If the robot adapts to changing conditions, then the recognition accuracy is maintained, but the task completion time increases due to model updating
Solution Approach 1:
The patent implements incremental model updating that processes only the necessary information from each recognition result and task outcome. Rather than performing complete model retraining that would consume excessive time, the system applies partial updates based on specific feedback signals, achieving adaptation with minimal time penalty while maintaining recognition accuracy
Data Source
AI summary
An object detecting method includes imaging a plurality of target objects with an imaging section and acquiring a first image, recognizing an object position/posture of one of the plurality of target objects based on the first image, counting the number of successfully recognized object positions/postures of the target object, outputting, based on the object position/posture of the target object, a signal for causing a holding section to hold the target object, calculating, as a task evaluation value, a result about whether the target object was successfully held, updating, based on an evaluation indicator including the number of successfully recognized object positions/postures and the task evaluation value, a model for estimating the evaluation indicator from an imaging position/posture of the imaging section and determining an updated imaging position/posture, acquiring a second image in the updated imaging position/posture, and recognizing the object position/posture of the target object based on the second image.