AGIBOT Releases WITA-Omni: Industry's First End-to-End Embodied Multimodal Model
AGIBOT has officially released WITA-Omni, hailed as the industry's first end-to-end multimodal embodied interactive large model. Built on streaming full-duplex architecture and trained on robot proprioceptive data, it integrates audio, video, tactile, and proprioceptive inputs with robotic actions and expressions. This model provides robots with a true interactive brain for the physical world, cutting down processing latency and eliminating the need for wake words. It marks a significant step forward in making human-robot interaction more natural, responsive, and context-aware. WITA-Omni achieves a fast interaction response time ranging from 400 to 600 milliseconds, roughly 3 times faster than the industry average. It also supports multi-language free communication, real-time visualization on facial screens, and proactive interaction triggered by environmental awareness.
## BACKGROUND
Embodied AI is an interdisciplinary cutting-edge field connecting artificial intelligence and robotics, where an intelligent agent possesses a physical body to interact with and learn from the real world. Full-duplex architecture allows a system to process inputs and generate outputs simultaneously without waiting for the input stream to finish.