PRISM: A Multi-View Multi-Capability Retail Video Dataset for Embodied Vision-Language Models Paper • 2603.29281 • Published Mar 31