LingBot-VA

LingBot-VA is a World-Action Model (WAM) developed by Robbyant Technology. It uses an autoregressive (AR) diffusion framework for simultaneous world modeling and action inference — architecturally unifying visual dynamics prediction and action inference within a single interleaved sequence while maintaining their conceptual distinction. Features a dual-stream mixture-of-transformers (MoT) architecture with asynchronous execution and KV caching for high-efficiency deployment. Demonstrates strong long-horizon performance, sample efficiency, and generalization to novel scenes. Published under Apache 2.0 license.