← 返回 Skills 市场

Habitat-GS-Navigator

Name: Habitat-GS-Navigator
Author: the0xka1

作者 The0xKa1 · GitHub ↗ · v0.1.1 · MIT-0

cross-platform ⚠ suspicious

213

总下载

当前安装

版本数

在 OpenClaw 中安装

/install habitat-gs-navigator

功能描述

Navigate and interact with photo-realistic 3DGS environments via the Habitat-GS Bridge. Use when: user asks to explore a 3D scene, perform embodied navigatio...

使用说明 (SKILL.md)

Habitat-GS Navigator

Control an embodied agent inside photo-realistic 3D Gaussian Splatting environments through the Habitat-GS Bridge.

Installation

git clone https://github.com/The0xKa1/habitat-gs-bridge.git
cd habitat-gs-bridge
pip install -e .

This provides two commands:

hab-cli — CLI for controlling the simulator (used by this skill)
habitat-gs-bridge — starts the bridge server

For full API details, read references/api-reference.md. For setup instructions, read references/setup.md.

Quick Workflow

# 1. Start the bridge server (in a separate terminal)
habitat-gs-bridge

# 2. Verify it's running
hab-cli status

# 3. Load scene (by scene-id + dataset, or by direct path)
hab-cli load_scene --scene-id gs_scene --dataset /path/to/config.json
hab-cli load_scene --scene /path/to/scene.gs.ply

# 4. Reset episode with start/goal
hab-cli reset --start "5.18,-3.57,-2.86" --goal "-3.62,-3.61,3.18"

# 5. Navigate: observe → decide → act → repeat
hab-cli step move_forward
hab-cli step turn_left
hab-cli step turn_right
hab-cli step stop          # when goal reached

# 6. Utilities
hab-cli observe             # current observation without stepping
hab-cli path --goal "x,y,z" # shortest-path info
hab-cli random_point        # sample navigable point

Navigation Loop

Observe: read agent_state.position, distance_to_goal, collided
Decide: use the philosophical-three-questions skill (Goal/State/Future tree)
Act: pick one of move_forward, turn_left, turn_right, stop
Check: verify distance decreased; if collided, turn to find open path
Repeat until done is true or distance_to_goal \x3C goal_radius

Decision Heuristics

collided after move_forward → turn (try left, then right) to find open path
distance_to_goal decreasing → keep current heading
distance_to_goal stagnant/increasing → change direction, use hab-cli path to check geodesic distance
distance_to_goal \x3C 0.5m → call stop
Near max_steps → consider stop if reasonably close

Configuration

The bridge server URL defaults to http://127.0.0.1:8890. Override with:

--url flag: hab-cli --url http://host:port status
Environment variable: export HABITAT_GS_BRIDGE_URL=http://host:port

Experience Logging

After each episode, record to ~/.openclaw/workspace/memory/YYYY-MM-DD.md:

## [NAV] Episode \x3Cid> in \x3Cscene>
- Result: success/fail (N steps, optimal: M steps)
- Key decisions: \x3Cturning points>
- Lesson: \x3Cwhat to do differently>

After 5+ episodes, review memory and extract recurring patterns into new skills or update this skill's heuristics.

安全使用建议

What to consider before installing and using this skill: - The skill is coherent with its description: it controls a local Habitat-GS bridge and uses hab-cli to navigate scenes. No credentials are requested. - The SKILL.md instructs you to git clone and pip install -e from a third‑party GitHub repo. That will run arbitrary code on your machine during install — review the repository contents (setup.py/pyproject, entry points, any scripts) before installing. - Prefer to install in an isolated environment (a dedicated conda env or virtualenv) or a disposable container, not as root or into your system Python, because habitat_sim and other native deps can be heavy and may require compilation. - Check the repo author and consider using an official release tag or checksum instead of cloning the main branch. If you cannot audit the code, run the service in a sandboxed VM or container. - Expect the skill to write episode logs under ~/.openclaw/workspace/memory/. If you have privacy concerns about local recordings or generated observations (the API returns base64 images), confirm what gets persisted and where. - If you plan to point the bridge at a remote host, be cautious: the CLI supports a --url / HABITAT_GS_BRIDGE_URL and could talk to remote servers; for normal use keep it on localhost. If you want a stronger safety posture: inspect the GitHub repository, prefer pinned releases, run the bridge in a container, and avoid installing into system Python environments.

功能分析

Type: OpenClaw Skill Name: habitat-gs-navigator Version: 0.1.1 The skill facilitates 3D environment navigation by installing and executing external code from a GitHub repository (https://github.com/The0xKa1/habitat-gs-bridge.git) and interacting with a local server via shell commands (hab-cli). While these capabilities—including shell execution, network access, and external dependency installation—are aligned with the stated purpose of interfacing with the Habitat-GS simulator, they represent high-risk behaviors according to the analysis criteria. No evidence of intentional malicious activity, such as data exfiltration or backdoors, was found in the provided files.

能力评估

✓ Purpose & Capability

Name/description (navigate 3DGS via Habitat-GS Bridge) align with the runtime instructions: the SKILL.md documents using hab-cli/habitat-gs-bridge, loading scenes, stepping the agent, and reading agent state. Required resources (local bridge, scene files, habitat_sim) are appropriate for the stated purpose.

ℹ Instruction Scope

Instructions stay within navigation/simulator scope (start bridge, use hab-cli to load/reset/step/observe, inspect distance/collision). The skill instructs recording episode summaries to ~/.openclaw/workspace/memory/YYYY-MM-DD.md (local logging), which is reasonable for a memory feature but is a persistent write to the user's home directory and should be expected by the user.

⚠ Install Mechanism

The SKILL.md tells the user to git clone and pip install -e from a GitHub user repo (The0xKa1). Installing arbitrary third‑party code with pip (editable install) executes code during install and can run arbitrary setup steps; the instructions do not pin a release, provide checksums, or indicate official/verified upstream. This is a legitimate install for this tool but is higher risk — run in an isolated environment and review the repo before installing.

✓ Credentials

No credentials or sensitive environment variables are requested. The docs optionally reference HABITAT_GS_BRIDGE_URL and BRIDGE_PORT to point the CLI at a non-default host/port, which is reasonable. The skill does require local access to scene files and habitat_sim (heavy native deps), which is appropriate for the task.

ℹ Persistence & Privilege

The skill does not request elevated privileges and is not always-on. It writes episode logs to ~/.openclaw/workspace/memory/, creating persistent local files (expected for memory features). Autonomous invocation is allowed (platform default) but not by itself a concern; this skill does not modify other skills or system-wide configurations.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install habitat-gs-navigator
安装完成后，直接呼叫该 Skill 的名称或使用 /habitat-gs-navigator 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v0.1.1

- Updated installation instructions to use a git clone and editable pip install instead of pip for latest habitat-gs-bridge. - No functional or API changes to skill usage or commands.

v0.1.0

- Initial release of habitat-gs-navigator skill for embodied navigation in photo-realistic 3DGS environments using the Habitat-GS Bridge. - Supports commands like scene loading, agent movement, observation, and pathfinding via the hab-cli tool. - Provides workflow and setup instructions for running the bridge server and interacting with 3DGS scenes. - Includes navigation loop and decision heuristics for agent behavior and episode management. - Offers configuration options for customizing the bridge server URL and experience logging for performance review.

元数据

Slug habitat-gs-navigator

版本 0.1.1

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 2

常见问题