World models · Embodied agents · Multimodal real-time companions · Learning from human–AI interaction · Latent action representations · Efficient on-device inference