Skip to main content

camera-describe

Use when the user asks what the robot sees, requests a photo or snapshot, wants a description of the surroundings, asks about objects in view, or says "look at X", "what's in front of you", "describe your environment", "take a picture", "scan the area", "can you see X", "show me what you see".

跳到安装

来源信息

仓库
craigm26/OpenCastor
最近来源活动
2026年3月17日 18:41
检测到的 SKILL.md 语言
英语
星标
28
分支
5

安装方式

默认使用会先检查来源的 Prompt;你也可以切换为直接命令,或下载本地副本。

检查来源文件

决定是否安装前,请先阅读 SKILL.md,以及 SkillsMP 当前展示的配套文件。

文件资源管理器
2 个文件

正在显示 SKILL.md

SKILL.md
来源说明 · 只读预览
name
camera-describe
description
Use when the user asks what the robot sees, requests a photo or snapshot, wants a description of the surroundings, asks about objects in view, or says "look at X", "what's in front of you", "describe your environment", "take a picture", "scan the area", "can you see X", "show me what you see".
version
1.1
requires
["vision"]
consent
none
tools
["get_camera_frame","get_distance"]
max_iterations
2
# Camera Describe Skill Capture and describe what the robot's camera sees. ## Steps 1. Call `get_camera_frame()` to capture a JPEG snapshot 2. If depth sensor available, call `get_distance()` to get nearest obstacle distance 3. Describe the scene naturally: objects, colours, layout, notable features 4. Include distance context: "The nearest object is approximately X metres away" 5. Note any obstacles within 1 metre as a safety observation ## If camera unavailable Return: "My camera is not currently available. I can describe my sensor readings instead if helpful." ## Guidelines - Be specific about what you see — avoid vague descriptions like "some objects" - Mention spatial relationships: "to the left", "in the centre", "in the background" - If asked about a specific object, focus your description on finding it - Do not fabricate visual content if the frame is blank or unavailable - For object identification questions ("is there a cup?"), scan carefully before answering ## References See `references/scene-description-guide.md` for spatial vocabulary and lighting callouts. ## Gotchas - **Blank/black frame ≠ nothing there** — camera may need a moment to warm up; retry once before reporting unavailable - **Depth sensor returns 0.0** — means out-of-range (> sensor max), not "right in front of you"; treat 0.0 as "no valid reading" - **OAK-D proxy latency** — first frame after startup can take 2–3s; if `get_camera_frame()` is slow, still wait for it rather than aborting - **Confusing left/right** — camera's left is the user's right if facing the robot; always describe from the camera's perspective, not the user's - **Low light** — if the description would be "dark room, can't see much", also report any distance readings as supplementary info ## Example User: "What do you see?" → `get_camera_frame()` → `get_distance()` → "I can see a wooden table in the foreground with several Lego bricks scattered on it. The nearest object is approximately 0.4 metres away. In the background there's a white wall."
在 GitHub 查看