メニュー
Skill for understanding RGB images from three cameras. Trigger when the user mentions "understand images" or "understand the environment".
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | understand-three-view-images |
| description | Skill for understanding RGB images from three cameras. Trigger when the user mentions "understand images" or "understand the environment". |
This skill understands images from the robot head, left-arm, and right-arm cameras. It auto-captures view images, runs a multimodal model, and returns an environment understanding result.
Before running, set ARK_API_KEY (or VOLCENGINE_API_KEY) and source $ZERO2SKILL_ROOT/configs/paths.env.
scripts/auto_understand_images_bbox.shUse scripts/auto_understand_images.sh --prompt "Please describe the current desktop scene" to get environment understanding. The script auto-captures images and calls the understanding script.
understand-three-view-images/
├── SKILL.md
└── scripts/
├── auto_understand_images.sh
├── auto_understand_images_bbox.sh
├── capture_and_understand_one_view.py
└── ...
scripts/auto_understand_images.sh: capture and understand; requires --promptFirst analyze the task and generate a collect/reset plan, then loop collect + reset until done. Trigger when the user mentions "analyze task", "step breakdown", or "learn a task".
Use when the user mentions "grasp", "pick up", or "pick and place". Uses SAM3 text segmentation + AnyGrasp planning to drive a robot arm for grasping, optionally place at a given (x,y); supports --collect to record HDF5.
Robot-arm self-learning skill that orchestrates analysis, collection, reset, and related skills in order.