Surface Detection Models for Robotic Lawnmower Radar

The patent divides the surface type detection task into multiple specialized machine learning models, each trained for a specific drive wheel rotational velocity range. This segmentation allows each model to specialize in detecting surface types under specific velocity conditions, thereby improving detection accuracy without requiring a single overly complex model to handle all scenarios.

The system dynamically selects the appropriate machine learning model based on the current drive wheel rotational velocity. By matching the operating conditions with the corresponding trained model, the system ensures optimal detection accuracy for each velocity regime while keeping individual model complexities manageable.

Instead of training one model to handle all possible velocity conditions, the patent trains multiple models each covering a specific velocity range. This partial action approach allows each model to be simpler and more accurate for its designated range, while the collection of models collectively covers the full operating spectrum.

The patent changes the training parameter of drive wheel rotational velocity across different models. Each model is trained with velocity-specific data, creating a set of models that collectively cover the full range of operating velocities. This parameter-based differentiation improves accuracy without requiring each individual model to be overly complex.

The machine learning models are pre-trained offline for specific velocity ranges, so that during runtime, the system only needs to select the appropriate pre-trained model and feed radar data into it. This preliminary training action shifts the computational burden from runtime to training time, improving reliability without significantly increasing real-time energy consumption.

The patent creates multiple copies of machine learning models, each specialized for a velocity range. Once trained, these model copies can be efficiently deployed and executed. The copying approach allows parallel optimization of each model for its specific purpose, improving reliability while enabling efficient runtime execution through specialized, optimized model instances.