http://localhost:1313/tags/vlm/ Recent content in VLM on Markus Knauer Hugo en-us Mon, 20 Apr 2026 00:00:00 +0000 http://localhost:1313/posts/clasp-language-driven-skill-selection-and-composition/ Mon, 20 Apr 2026 00:00:00 +0000 http://localhost:1313/posts/clasp-language-driven-skill-selection-and-composition/ <p>How can a robot understand <em>what</em> you want — and admit <em>when it doesn&rsquo;t know how</em>? Our framework <strong>CLASP</strong> bridges the gap between data-hungry vision-language-action models and data-efficient imitation learning by combining <strong>Vision-Language Models</strong> with <strong>Task-Parameterized Kernelized Movement Primitives (TP-KMPs)</strong>, requiring only <strong>2–5 demonstrations per skill</strong>.</p> <picture class=""> <source srcset="http://localhost:1313/images/clasp_overview.webp" type="image/webp"> <img src="http://localhost:1313/images/clasp_overview.png" alt="CLASP execution pipeline: the VLM matches a natural language query against the skill library, composes existing skills when possible, and requests new demonstrations when a capability gap is detected" loading="lazy"> </picture> <p>CLASP operates in three stages:</p>