Robbyant Releases Full-Stack Embodied AI Models for Robot Deployment

Robbyant, an embodied AI company within Ant Group, completed a launch week by releasing a six-model full-stack of AI tools designed to bridge the gap between digital simulation and physical robot control. The rollout culminated on July 10 with the debut of LingBot-VA 2.0, the first embodied-native video-action world model built specifically for physical world control. This model utilizes an autoregressive architecture to achieve an inference speed of 150 Hz on a single GPU.

Earlier in the week, the company open-sourced LingBot-World 2.0 (Infinity), an interactive world model capable of hour-long real-time generation at 720p/60fps. To prevent visual degradation over long horizons, the model employs a Causal Pretraining Paradigm and a proprietary Mask of Bidirectional Attention mechanism. Robbyant also released LingBot-Video, a video generation foundation model based on a Mixture-of-Experts architecture.

These releases aim to facilitate robot deployment in sectors such as elderly care and medical assistance. CEO Zhu Xingliang stated that the company will continue to explore limits in embodied intelligence while developing an open technology ecosystem to expedite the use of robots in real-world and industrial scenarios.