Lightweight Robotic Brain: Offline Natural Language Interaction and Decision-Making
Project Description
This project aims to develop a lightweight "AI brain" based on large language models (LLMs) for offline deployment. Focusing on real robotic platforms with limited computing power and no reliance on cloud services, we will employ advanced model compression and inference optimization techniques. Our goal is to utilize natural language dialogue—the most intuitive form of interaction—to enable real-time understanding of complex instructions, contextual reasoning, and the generation of reliable, efficient robot control code and decision sequences. Ultimately, this will allow complex tasks to be executed through natural language commands.
Supervisor
SHI, Ling
Quota
1
Course type
UROP1100
UROP2100
UROP3100
UROP3200
UROP4100
Applicant's Roles
• Programming Skills.
• Priority given to those with experience in deploying and fine-tuning AI Agents.
• Priority given to those with experience in embedded system development.
• Familiarity with natural language processing.
• Priority given to those with experience in deploying and fine-tuning AI Agents.
• Priority given to those with experience in embedded system development.
• Familiarity with natural language processing.
Applicant's Learning Objectives
(1) Master lightweight techniques, knowledge distillation, and efficient inference methods for deploying LLM on resource-constrained platforms.
(2) Deploy a locally lightweight LLM to enable natural language interaction.
(3) Generate reliable and efficient robot control code and decision sequences through the locally deployed LLM.
(2) Deploy a locally lightweight LLM to enable natural language interaction.
(3) Generate reliable and efficient robot control code and decision sequences through the locally deployed LLM.
Complexity of the project
Challenging