Robotic Visual Output Correction for Drivable Area Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robotic systems rely on retraining neural networks to correct visual output errors, which is inefficient, and their object detection robustness is limited to unique features rather than pixel-level comparisons.
Innovation Solution
The system introduces a method for visualizing data generated by robotic devices, allowing operators to identify and correct drivable areas without retraining the neural network, by comparing pixel descriptors between current images and keyframes, and updating parameters based on relative transforms, enabling more accurate path determination and task execution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional systems retrain neural networks to correct visual output errors, then the neural network can learn from new data, but the process is inefficient and time-consuming
Solution Approach 1:
The patent creates a virtual copy of the real-world environment through captured images and generates a virtual drivable area in this digital replica. Operators can annotate and correct the virtual drivable area without affecting the actual robotic device, enabling rapid iteration and correction without physical retraining time
Solution Approach 2:
The system introduces a virtual environment as an intermediary between the neural network and the real world. Corrections are first made in the virtual space through operator annotations, then transferred to update the neural network, avoiding direct and time-consuming retraining loops
2Reliability
If conventional systems use unique features for object detection, then the detection process is computationally simpler, but the robustness is limited and cannot handle varying environments well
Solution Approach 1:
The patent transforms the detection approach from feature-based to pixel-based by changing the fundamental parameter of comparison. Instead of extracting and comparing unique features, the system compares pixel descriptors across entire images, enabling robust detection of drivable areas through detailed pixel-level analysis
Solution Approach 2:
The system adds a new dimension to object detection by moving from feature-space comparison to pixel-space comparison. This dimensional shift allows the neural network to detect drivable areas through pixel descriptor matching, providing greater robustness to environmental variations
3Adaptability or versatility
If the robotic device operates in varying environments, then the robot's versatility increases, but the accuracy of path determination decreases without retraining
Solution Approach 1:
The system implements a feedback loop where the robotic device captures images from varying environments, operators annotate the virtual drivable area, and these corrections are used to update the neural network. This continuous feedback mechanism maintains accuracy across diverse environments without requiring extensive retraining
Solution Approach 2:
The system performs preliminary corrections in the virtual environment before deploying updated parameters to the robotic device. By pre-processing and annotating virtual images, the system prepares corrected data in advance, allowing the robot to maintain high accuracy when operating in new environments
Data Source
AI summary
A method for visualizing data generated by a robotic device is presented. The method includes displaying an intended path of the robotic device in an environment. The method also includes displaying a first area in the environment identified as drivable for the robotic device. The method further includes receiving an input to identify a second area in the environment as drivable and transmitting the second area to the robotic device.