Robotic Mobile Recognition Model Updates with Virtual Space Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Recognition models in robotic mobile devices, such as robot vacuum cleaners, exhibit varying performance due to the type of object being recognized and the surrounding environment, leading to issues like incorrect object recognition and operational failures.
Innovation Solution
A method involving an electronic device that obtains spatial scan data from the robotic mobile device, processes it to extract spatial information and virtual object data using a generative model, and updates the recognition model with training data generated from a virtual space simulating the real environment to improve recognition performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a recognition model is used in robotic mobile devices, then object recognition capability is provided, but recognition performance varies due to different spaces and environments
Solution Approach 1:
The patent applies local quality by creating space-specific recognition models tailored to individual environments. Each space receives customized training data and model parameters based on its unique characteristics (furniture types, layouts, lighting conditions), allowing the recognition model to optimize performance for that specific location rather than using a generic model for all spaces.
Solution Approach 2:
The patent implements parameter changes by dynamically updating model parameters based on spatial information extracted from each environment. The system adjusts recognition parameters according to detected space characteristics, object distributions, and environmental factors, enabling the model to adapt its behavior and performance parameters to match the specific requirements of different spaces.
2Measurement precision
If training data is collected from real spaces, then recognition accuracy improves, but data collection time and computational resources increase
Solution Approach 1:
The patent applies copying by creating virtual representations of real spaces through 3D modeling and simulation. Instead of collecting training data directly from physical spaces, the system generates synthetic training data from virtual copies of the environments, preserving the essential spatial relationships and object characteristics while eliminating the time-consuming process of real-world data collection.
Solution Approach 2:
The patent implements preliminary action by pre-processing and pre-generating training data in virtual environments before actual model training occurs. The system performs preliminary 3D mapping, object detection, and data generation in silico, so that when real space data is needed, the foundation is already prepared and only fine-tuning is required.
3Reliability
If the recognition model is updated with space-specific data, then performance in specific spaces improves, but model complexity and processing requirements increase
Solution Approach 1:
The patent applies segmentation by dividing the recognition system into modular components: a universal base model and multiple space-specific refinement models. Each space-specific model handles only the unique characteristics of its target environment, while the base model provides common functionality, reducing overall complexity compared to creating entirely separate models for each space.
Solution Approach 2:
The patent uses an intermediary processing layer that translates real space data into virtual representations and back into optimized model parameters. This intermediary virtual environment acts as a mediator between raw sensor data and final model updates, simplifying the processing requirements by performing complex transformations in a controlled virtual domain rather than directly on real-world data.
Data Source
AI summary
A method of updating a recognition model of a robotic mobile device, the method including: obtaining, by an electronic device, from the robotic mobile device, spatial scan data regarding a target space; obtaining, by the electronic device, based on the spatial scan data, spatial information including information about a structure of the target space and an item in the target space; obtaining, by the electronic device, virtual object data including information about a class of a virtual object and a position of the virtual object by inputting the spatial information to a generative model; obtaining, by the electronic device, training data by using the spatial information and the virtual object data; and updating, by the electronic device, the recognition model of the robotic mobile device using the training data.