【#Tech24H】On October 29, AgiBot released WITA-OMNI, the industry's first embodied omnimodal end-to-end interaction large model, delivering a host of core technical upgrades in perception, interaction and autonomous decision-making for robots. It is reported that WITA-OMNI adopts a full-duplex streaming link architecture. With in-depth optimization, the robot achieves a fastest response latency of 400 milliseconds and an average latency of around 600 milliseconds — over three times faster than mainstream industry solutions. It eliminates the lag commonly seen in interactions with traditional large models and delivers a conversational rhythm close to that of humans. The model boasts a set of core capabilities: enhanced sound capture, which leverages audio-visual collaborative perception to stably pick up target speech in noisy multi-person scenarios; omnimodal out-of-scope rejection, which combines sound, lip movement, semantics and context for joint understanding and enables wake-word-free interaction at all times; full-time autonomy, allowing independent execution of perception, decision-making, action and continuous interaction; ultra-low latency response supported by streaming links and optimized inference. It also supports seamless switching across more than 100 languages and 29 dialects, with functions for simultaneous interpretation and bilingual translation.