Agibot released WITA-Omni, an embodied omni-modal model with response times as fast as 400 milliseconds
Agibot released WITA-Omni on November 10, described as the first embodied omni-modal end-to-end interaction model, using a streaming full-duplex architecture to listen, think, respond and act simultaneously with response times as fast as 400 milliseconds and averaging 600 milliseconds.
Agibot announced through its official account on the evening of November 10 that it had released WITA-Omni, a silicon-photonics language omni-modal large model. According to the official introduction, it is the industry first embodied omni-modal end-to-end interaction large model. WITA-Omni is deeply trained on the robot own-body perception data and, in addition to audio and video input and output, newly supports tactile and proprioceptive state input as well as robot motion and facial expression output, giving robots an interaction brain built for the physical world. Based on a streaming full-duplex architecture, WITA-Omni can listen, think, respond, act and make expressions simultaneously, reducing the waiting inherent in traditional serial processing, with interaction responses as fast as about 400 milliseconds and averaging about 600 milliseconds, nearly 3 times faster than the industry average. Omni-modal rejection removes the need for wake words, allowing fully natural conversation, with the system combining conversation orientation, video, audio, voiceprint, lip movement, posture, semantics and context to decide whether the current utterance needs a response, greatly reducing false triggers during multi-party conversation or ambient broadcast interference. The model supports free switching-free conversation across more than a hundred languages and dialects and, combined with the full-duplex architecture, achieves streaming simultaneous interpretation. When content is hard to convey in words, such as explaining science to a child, WITA-Omni intelligently decides to visualize it in real time on a facial display. The model can also continuously perceive the surrounding environment and user state, think and decide autonomously, and proactively initiate interaction at appropriate moments. The company previously disclosed that WITA-Omni Preview topped the DailyOmni omni-modal understanding leaderboard with a composite score of 85.21.