sii-rhos-ai/ViFailback-8B
ViFailback-8B is an 8 billion parameter Vision-Language Model (VLM) developed by sii-rhos-ai, fine-tuned from Qwen3-VL-8B-Instruct with a 32K context length. It is specifically designed for diagnosing robotic manipulation failures and generating both textual and visual correction guidance. The model uniquely employs explicit visual symbols like arrows and crosshairs to facilitate learning and recovery from real-world robotic failures.
Loading preview...
ViFailback-8B Overview
ViFailback-8B is an 8 billion parameter Vision-Language Model (VLM) developed by sii-rhos-ai, fine-tuned from Qwen3-VL-8B-Instruct. This model is specifically engineered to address robotic manipulation failures by providing comprehensive diagnostic and corrective guidance.
Key Capabilities
- Robotic Failure Diagnosis: Identifies and analyzes the root causes of manipulation failures in robotic systems.
- Textual Correction Guidance: Generates natural language instructions to help rectify identified issues.
- Visual Correction Guidance: Utilizes explicit visual symbols, such as arrows, crosshairs, and state icons, to visually communicate corrective actions. This unique feature helps bridge the gap between abstract failure diagnosis and concrete policy correction, enabling robots to learn and recover more effectively from real-world scenarios.
Unique Approach
The model's core innovation lies in its use of visual symbols to provide actionable feedback, allowing robotic systems to interpret and implement corrections more intuitively. This approach is detailed in the associated research paper: Diagnose, Correct, and Learn from Manipulation Failures via Visual Symbols.
Usage
ViFailback-8B can be used for inference to render corrective visual symbols, with usage examples provided in its official GitHub repository.