Encord releases EBIND multimodal embedding model for AI agents - The Robot Report

Encord, a London-based company specializing in data infrastructure for physical and multimodal AI, has launched EBIND, a new embedding model designed to enhance AI agents, robots, and other systems that process multimodal data. Built on Encord’s recently released E-MM1 dataset—claimed to be the largest open-source multimodal dataset globally—EBIND supports five data modalities, including audio, video, text, images, and 3D point clouds from lidar sensors. This enables AI models to interpret complex environments by integrating diverse data types, such as extrapolating 3D shapes from 2D images or locating video content based on voice prompts, thereby improving the understanding of objects and their spatial relationships. The E-MM1 dataset was painstakingly constructed due to the rarity of comprehensive multimodal data quintuples in the wild, making it 100 times larger than the next largest dataset. EBIND’s high-quality data foundation allows the model to be smaller, faster, and more cost-effective than