ET_Robot Project

This repository contains the necessary scripts, datasets, and configurations for training, fine-tuning, and evaluating robot behavior models using various tools and datasets.

Directory Structure

• box_evaluation/:
  Folder for evaluating performance and accuracy of box-related tasks or object manipulation.

• phi-3-mini-LoRA/:
  Directory for the low-rank adaptation (LoRA) models used in fine-tuning the robot models with a focus on performance and efficiency.

• prompts_folder/:
  Includes various prompts used during training and fine-tuning to guide the model's behavior.

• robosuite/:
  This folder includes the Robosuite framework for creating simulation environments and tasks for the robot. Robosuite is likely used here for simulating the tasks.

Scripts and Files

• data_generation.py:
  Script for generating finetuning data to train phi-3-mini models.

• data_prep_stack_three.txt & data_prep_stack_two.txt:
  Configuration or instruction files for preparing data specific to the stacking tasks. Likely contains preprocessing steps for different types of data stacks.

• finetuning.py:
  Script for fine-tuning pre-trained models on specific tasks.

• main_retrieval_GPT-4o.py:
  Script for retrieving information and generating tasks using GPT-4-based models in conjunction with the robot’s system.

• stack_multiple_record.py:
  Script for managing and recording results from multiple stack trials or tasks.

• updated_val_data.json:
  JSON file containing validation data used during model training and evaluation phases.

How to Run

1. Install necessary dependencies:

   pip install -r requirements.txt
   

2. Install necessary dependencies from robosuite: https://robosuite.ai/