July 13, 2026

Robbyant unveils LingBot-VA 2.0, a real-time robot control model

The new LingBot-VA 2.0, developed by Robbyant, revolutionizes robot control by training natively for physical actions instead of adapting traditional content-creation video generators. This model distinguishes itself by implementing closed-loop control that predicts future actions while executing them, achieving a peak asynchronous execution frequency of 225 Hz. Unlike most video-action models that prioritize visual appearance, LingBot-VA 2.0 utilizes self-supervised learning from unlabeled web videos, emphasizing causal prediction and action-relevant features. This innovative approach allows it to adapt efficiently across different robot embodiments, demonstrating superior performance in real-world tasks compared to previous models.

Robbyant: Robbyant is an embodied AI company operating under Ant Group. It developed LingBot-VA 2.0 as a native video-action foundation model optimized for real-time robot control. The company focuses on rebuilding tokenizers and backbones around physical dynamics rather than content creation.
Ant Group: Ant Group is the parent organization of Robbyant, the embodied AI company. It backs the development of LingBot-VA 2.0 to advance responsive robotic systems through native training for action and control.
LingBot-VA 2.0: LingBot-VA 2.0 is a video-action foundation model designed from the ground up for robot control. It incorporates a causal policy, sparse Mixture-of-Experts architecture, foresight reasoning for ongoing prediction, and multi-chunk training to capture motion dynamics. The model supports adaptation from limited demonstrations and transfer across different robot embodiments.

Model Training: LingBot-VA 2.0 trains natively on physical control objectives using self-supervised dynamics from unlabeled web video instead of adapting existing video generators.
Embodied AI Focus: Robotic policies benefit from specialized tokenizers and backbones that prioritize action-relevant features and causal forward prediction over visual reconstruction.
Real-time Execution: The architecture enables closed-loop control where predictions run concurrently with actions and are continually updated by fresh observations.

AI tools built by Emerald Force

Built and supported by Emerald Force.

You might also like