robotwin-image-to-env

What problem does it solve?

This skill converts a single image into a reviewed RoboTwin environment by detecting visible tabletop objects, matching assets from a built-in library, localizing anchors on the tabletop with locate MCP, drafting the layout, and generating a renderable first frame.

Core Features & Use Cases

• Object discovery on the tabletop using qwen3-vl-plus, followed by per-object cropping.
• Asset candidate retrieval from the built-in multimodal asset library via qwen3-vl-embedding, with instance-level or class-level matching.
• Pose preset selection per object using asset docs, then authoritative localization via locate-tabletop-contacts MCP.
• Estimation of a conservative tabletop footprint and projection into a RoboTwin layout, followed by a render pass and VLM-based refinement.
• Honest reporting for unresolved objects and generation of a draft RoboTwin env for user review.

Quick Start

Provide a source image; the system will generate a draft RoboTwin env and a first render for review.