ViFailback-8B

ViFailback-8B

ViFailback-8B is a Vision-Language Model (VLM) designed to diagnose robotic manipulation failures and provide both textual and visual correction guidance. It is fine-tuned from Qwen3-VL-8B-Instruct as part of the ViFailback framework.

The model utilizes explicit visual symbols (arrows, crosshairs, state icons) to bridge the gap between failure diagnosis and policy correction, allowing robotic systems to learn from and recover from real-world failures.

• Paper: Diagnose, Correct, and Learn from Manipulation Failures via Visual Symbols
• Project Page: https://x1nyuzhou.github.io/vifailback.github.io/
• Repository: https://github.com/x1nyuzhou/ViFailback

Usage

To run inference and render the corrective visual symbols, use the vifailback_infer.py script provided in the official GitHub repository:

python vifailback_infer.py \
    --model_path sii-rhos-ai/ViFailback-8B \
    --json_path ./examples/example_vifailback_infer.json \
    --dataset_root /path/to/ViFailback-Dataset \
    --output_dir ./inference_visualizations

Citation

@inproceedings{zeng2026diagnose,
  title={Diagnose, correct, and learn from manipulation failures via visual symbols},
  author={Zeng, Xianchao and Zhou, Xinyu and Li, Youcheng and Shi, Jiayou and Li, Tianle and Chen, Liangming and Ren, Lei and Li, Yong-Lu},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={42386--42395},
  year={2026}
}