tnoinkwms/FastMotionGeneration

FastMotionGeneration

Introduction

This repository contains the implementation described in the paper From Text to Motion: Grounding GPT-4 in a Humanoid Robot “Alter3”.

Method

GPT already understands how humans move and feel. We use two prompts for this purpose. In prompt-1, the input command is described in about ten sentences. In prompt-2, each sentence is converted into Python code (motion signals). It is very simple, but it is enough to control a very complex robot with 43 joints!!

Request

To run this implementation, ensure you have the following dependencies:

openai == 1.35.13
langchain == 0.2.7

Do NOT forget replace the API with your own. (line53)

To generate code, use the following command:

python FastGen.py "Action"

Example:

python FastGen.py "pretend to be a ghost"

Note: You can generate the code, but you will encounter an error if not connected to the Alter3.