An integrated cognitive framework for Embodied AI that bridges visual perception, action planning, and self-evolutionary learning.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install vision-action-evolution-loop
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install vision-action-evolution-loop using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Vision-Action-Evolution Loop?
The Vision-Action-Evolution Loop is a sophisticated framework designed to create a closed-loop cognitive system for autonomous agents and robotics. It synthesizes two core philosophies: diepre-vision-cognition for robust sensory perception and self-evolution-cognition for continuous learning. By moving beyond simple end-to-end models, this skill within the Openclaw Skills library provides a structured bridge from 2D visual data to 3D spatial understanding and physical execution.
This framework implements the SOUL Five Laws—Perceive, Plan, Execute, Evaluate, and Evolve. It effectively wraps traditional high-precision pipelines like OpenCV and CAD vectorization into tool-augmented VLLM workflows, allowing agents to perform complex tasks such as robotic manipulation, folding, and assembly with increasing accuracy over time.
Vision-Action-Evolution Loop Use Cases
- Autonomous robotic arm manipulation and assembly based on visual feedback.
- Automated conversion of 2D photographs into 3D parametric CAD models.
- Quality control systems that utilize human-in-the-loop evaluation to refine AI decision-making.
- Development of self-improving Embodied AI agents that update their world models through task repetition.
How Vision-Action-Evolution Loop Works
- Perception: The system processes 2D inputs through perspective correction and line detection to generate 3D spatial feature maps.
- Planning: Utilizing VLA (Vision-Language-Action) models, the skill calculates grasp points and action sequences based on the spatial model.
- Execution: Physical or simulated actions are performed, such as grasping, folding, or moving objects.
- Evaluation: The system performs a visual re-inspection or requests human feedback to determine if the execution met the success criteria.
- Evolution: Successes and failures are recorded to update the internal cognitive nodes and refine future action parameters.
Vision-Action-Evolution Loop Setup
To install this skill using the CLI, run the following command:
clawhub install vision-action-evolution-loop
Alternatively, you can manually install the skill by copying the directory to your local configuration:
cp -r skills/vision-action-evolution-loop ~/.openclaw/skills/
Vision-Action-Evolution Loop Data Schema & Taxonomy
The skill organizes its cognitive lifecycle through a structured data output containing five distinct stages:
| Stage | Output Type | Description |
|---|---|---|
| Perception | Feature Map | 2D/3D spatial coordinates and confidence scores |
| Planning | Action Sequence | A structured list of robotic or tool-based commands |
| Execution | Physical State | Logs of the physical actions performed |
| Evaluation | Feedback | Pass/Fail status with re-inspection metadata |
| Evolution | Knowledge Node | Updated world model parameters and memory files |
Vision-Action-Evolution Loop Advanced Features
- Tool-Augmented VLLM Strategy: Wraps OpenCV and vectorization tools as callable functions for the model.
- Self-Evolving Embodied AI: Automatically updates memory and task logic based on previous execution results.
- Multi-Agent Support: Compatible with SAGE protocols for collaborative multi-agent evolution.
- 3D Parametric Mapping: Converts raw visual data into precise SVG, DXF, or 3D coordinate systems using state-of-the-art academic frameworks.
SKILL.md
Loading
Related Openclaw Skills
Academic Deep Search
A high-precision academic research skill designed to synthesize methods, markers, and findings from peer-reviewed literature into structured, evidence-based reports.
Mapbox Search Patterns
A specialized guidance skill for AI agents to master Mapbox search tools through optimized tool selection, parameter tuning, and spatial constraints.
Mapbox Search Integration
A complete workflow for implementing Mapbox search, guiding developers from initial discovery questions to production-ready application integration.
Mac Mini AI
A high-performance local AI orchestration layer that turns Mac Minis into scalable inference nodes for LLMs, image generation, and speech processing.
MapLibre to Mapbox Migration
A professional developer guide for transitioning web mapping applications from MapLibre GL JS to the Mapbox GL JS ecosystem.
FastapiAdmin WSL2 Deployment Skill
A comprehensive automation skill for deploying the FastapiAdmin stack within a Windows WSL2 Ubuntu environment.
Featured*
