Hierarchical LLMs In-the-loop Optimization

A hierarchical Large Language Models (LLMs) framework for real-time multi-robot task allocation and target tracking with unknown hazards.

About

Authors: Yuwei Wu, Yuezhan Tao, Peihan Li, Guangyao Shi, Gaurav S. Sukhatmem, and Vijay Kumar, and Lifeng Zhou

Video Links: Youtube

Related Paper: Yuwei Wu, Yuezhan Tao, Peihan Li, Guangyao Shi, Gaurav S. Sukhatme, Vijay Kumar, Lifeng Zhou, "Hierarchical LLMs In-the-loop Optimization for Real-time Multi-Robot Target Tracking under Unknown Hazards". 2024

System Architecture:

Prerequisites

• ROS: Our framework has been tested in ROS Noetic.

• Forces Pro: You can request an academic license from here.

• openai: Install by

pip install openai

Config

To check more scenarios and experiment settings, please refer to the files in "src/tracker/config/".

• LLM related paramters. Set true to run LLMs. The duration is to control the calling frequency. (will change to callback later)

llm_inner_dur: 2
llm_outer_dur: 10
llm_on: True

• Use steps and dt to control the experiment duration.

steps: 100
Problem: dt: 0.2

• Use task ability to change the maximum number of targets a robot can track

task_ability: 1

• The initial task assignment is given for the setup.

Run (The complete code will be released later)

python tracker_server.py exp1

change exp to test different settings.

• You may need to clean the solver folder when you change some setups from the problem and danger zone.

Maintaince

For any technical issues, please contact Yuwei Wu (yuweiwu@seas.upenn.edu).