US20260310206
2026-10-08
Physics
G05D1/2465
The patent describes a method for generating a spatial map that incorporates actual images captured from a space. This process begins with obtaining an initial map of the target area. A captured image of a specific area within this space is then registered to the initial map using depth information. This registration process helps integrate the actual image into the map, enhancing its realism and accuracy.
Traditionally, spatial maps are generated using LiDAR scan data, which can misrepresent the actual structure of a space due to the presence of various objects like furniture. LiDAR sensors might incorrectly identify these objects as part of the structural walls, leading to discrepancies between the generated map and the real-world space. This method aims to address such issues by incorporating actual images into the spatial map, thus providing a more realistic representation.
Once an actual image is registered to the initial map, the method identifies areas within the map that require image generation. Using a generative model, the system produces images for these areas, completing the spatial map. This comprehensive map reflects the true appearance and layout of the space, overcoming limitations of traditional LiDAR-based maps.
An electronic device, such as a robot cleaner, can implement this method. The device includes memory and processors that execute instructions to perform the mapping process. By capturing images and using depth information, the device integrates actual images with the initial map, enhancing its accuracy and visual fidelity.
This technology provides a significant improvement in spatial mapping by combining traditional LiDAR data with actual images. It enhances user perception by offering a more realistic map, useful in applications like home automation and robotics. The method ensures that maps are not only structurally accurate but visually representative of real-world environments.