Fusion of multiple text prompts into cosine map, SAM for segmentation + point cloud extraction and filtering to determine object stop radius more precisely

[naokiyokoyamabd]

• Value maps now support multiple channels for considering multiple text prompts
• Segment Anything Model used to segment object and extract its point cloud, which is post-processed / filtered using the Statistical Outlier Removal algorithm to remove noise. Since success is based on distance to outer surface of the object, this improves success rate significantly over the previous, much noisier method