CogExplore: Language Model-Guided Planning Informed by Prior States
Public Deposited- Abstract
The integration of language models into robotics enhances executive decision-making capabilitiesand establishes a foundation for world model grounding, opening new avenues in robotic navigation. Advanced models like GPT, Mistral and Gemini enable robots to make well-informed decisions by processing data from Visual Question Answering (VQA) models and real-time object detectors, leveraging their capabilities for cutting-edge exploration. Moreover, these models can compress observations and actions into natural language, forming a memory that guides future navigation. We propose a comprehensive pipeline that employs a ROS-Unreal interface to evaluate a language model-driven frontier-point exploration strategy. This approach showcases intuitive reasoning, leverages past experiences for future navigation, and generates a human-readable record of the decision-making process.
- Creator
- Date Issued
- 2024-04-22
- Academic Affiliation
- Advisor
- Committee Member
- Degree Grantor
- Degree Level
- Commencement Year
- Subject
- Publisher
- Last Modified
- 2025-01-07
- Resource Type
- Rights Statement
- Language
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Cooper_colorado_0051N_18896.pdf | 2024-12-13 | Public | Download |
|
|
Thesis_Approval_Form.pdf | 2024-12-13 | Public | Download |