Publication
Our paper, titled DKPROMPT: Domain Knowledge Prompting Vision-Language Models for Open-World Planning, was accepted at the AAAI LM4Plan Workshop 2025 and previously presented at CVPR EAI Workshop 2024.
This work introduces a novel approach to enhance vision-language models with domain-specific knowledge for improved robot planning in open-world scenarios.
Research Overview
Vision-language models show great promise for robot planning, but often lack domain-specific knowledge crucial for real-world applications. DKPROMPT addresses this limitation by incorporating structured domain knowledge into the prompting process.
Key Features
- Domain knowledge integration framework
- Enhanced vision-language model performance
- Open-world planning capabilities
- Comprehensive experimental validation
Impact
This research contributes to making robots more capable of understanding and operating in complex, real-world environments by bridging the gap between general-purpose models and domain-specific requirements.
Authors: Xiaohan Zhang, Zainab Altaweel, Yohei Hayamizu, Yan Ding, Saeid Amiri, Hao Yang, Andy Kaminski, Chad Esselink, and Shiqi Zhang (*Equal contribution)
Venues:
- AAAI LM4Plan Workshop 2025
- CVPR EAI Workshop 2024
| Paper | Project Website | arXiv |