Symmetry-Informed LLM/VLM + Deterministic Control for Robotic Object Manipulation (Educational DOFBOT)

🎥 Video

Click the image to watch on YouTube.

📘 Read this in Spanish / Leer en español

---

📚 Citation

If you use or reference this repository, please cite the published paper:

Gudiño-Lau, J., Durán-Fonseca, M., Anido-Rifón, L. E., & Santana-Mancilla, P. C.
A Symmetry-Informed Multimodal LLM-Driven Approach to Robotic Object Manipulation: Lowering Entry Barriers in Mechatronics Education.
Symmetry, 17(10), 1756. https://doi.org/10.3390/sym17101756

---

This repository provides a reproducible hybrid framework that integrates a multimodal Large Language Model (LLM/VLM) as a semantic reasoning module with deterministic inverse kinematics for robotic object manipulation.
The system is implemented on the Yahboom DOFBOT educational arm and demonstrates how symmetry can be applied as a design principle rather than a learned feature.

Designed for educational and research purposes, this project aims to lower entry barriers in mechatronics education by showing how generative AI can interact with deterministic robotic control.

---

🧩 Overview

Robot Side (MainRobot.py)

• Initializes and homes the robotic arm.
• Captures a camera image and exposes it through two TCP servers:
  • Image Server on port 6103.
  • Command Server on port 6104.

Client Side (RuedaColorEs.py)

• Connects to the robot’s image server to retrieve the latest frame.
• Sends both text and image to an OpenAI multimodal model (gpt-4o).
• Interprets the model’s reply (e.g., “lower right”) and sends the corresponding motion commands to the robot via TCP.

---

⚙️ Requirements

Hardware

• Yahboom DOFBOT 6-DoF robotic arm (or compatible)
• USB or CSI camera
• Local network connection between robot and client

⚖️ License

This project is released under the MIT License, allowing free use, modification, and redistribution for educational and research purposes.
See the LICENSE file for details.