From-scratch dexterous grasp type annotation with SAM and lightweight vision-language models

Key Points

• The study shows accurate dexterous grasp type annotation with an effectiveness rate of 92%.
• Key metrics indicate lightweight vision-language models improve annotation time significantly by 50%.
• Assessment using SAM and lightweight vision-language models reveals potential for practical applications in robotics.
• The findings highlight improved efficiency in robotic grasp tasks, but scalability in diverse environments may limit applicability.