TwelveLabs launches Pegasus 1.6 for physical AI video understanding

TwelveLabs released Pegasus 1.6, a model that understands egocentric video shot from the operator's point of view and turns it into structured training data for robots, drones and autonomous vehicles. It supports five workflows: action segmentation, dense captioning, quality scoring, search and consent flagging.

• Pegasus 1.6 is TwelveLabs' first model built for egocentric video
• Supports five workflows: action labeling, captioning, quality scoring, search, compliance
• No proprietary cameras required; works with existing video data
• Improved entity recognition for hands, objects and tools across clips