https://mrtanke.github.io/tags/vision-action-model/ Recent content in Vision-Action-Model on Tan Ke Hugo -- 0.160.1 en-us Mon, 16 Mar 2026 20:21:33 +0000 https://mrtanke.github.io/posts/2026-03-16-robbyant-work/ Mon, 16 Mar 2026 20:21:33 +0000 https://mrtanke.github.io/posts/2026-03-16-robbyant-work/ <p><a href="https://technology.robbyant.com/" target="_blank" rel="noopener noreferrer"><strong>Robbyant</strong></a> is a company under <a href="https://www.antgroup.com/" target="_blank" rel="noopener noreferrer">Ant Group</a>, dedicated to building the foundational platform for Embodied AI, bridging the gap between digital intelligence and the physical world.</p> <p>Since the company is still relatively new, I want to quickly review its recent work. In particular, I will study four embodied intelligence model models: <strong>spatial perception model, VLA model, world model, and video action model</strong>.</p> <p><img alt="image.png" loading="lazy" src="https://mrtanke.github.io/posts/2026-03-16-robbyant-work/image.png"></p> <p>This diagram in the homepage of Robbyant reflects the <strong>vision for embodied intelligence</strong>: starting from <strong>sensory input</strong>, the system first builds <strong>spatial intelligence</strong> to understand the physical world, then relies on an <strong>action model</strong> to make decisions and interact with the environment, and finally improves through <strong>environmental reward</strong>.</p>