Project VULCAN

Project VULCAN is an open humanoid robotics platform focused on expressive human-robot interaction through modular mechanical, electronic, and AI-driven subsystems. It is a humanoid robot facial structure that can interact with humans — tracking faces, expressing emotions, hearing and speaking.

Current Status

Actively undergoing infrastructure modernization and subsystem stabilization before next-generation expansion.

Architecture Principles

• Atomic modular subsystems.
• Minimal cross-module interference.
• Hardware/software separation.
• Scalable pipelines for future contributors.
• Experiment isolation via experiments/.

System Pipeline Overview

 Microphone Input
        ↓
 Speech-to-Text
        ↓
 Emotion + Context Fusion
        ↓
 LLM / RAG Pipeline
        ↓
 Response Generation
        ↓
 TTS + Mouth Movement + Facial Expressions

Features

• Eyes — Independent 2-axis eyeball motion, eyelids, and gaze tracking via embedded cameras.
• Mouth — Lip movement synced with speech and emotional expressions.
• Eyebrows — Up/down motion to complement expressions.
• Neck — 2-axis motion (horizontal 360° + vertical tilt).
• Hearing & Speech — Microphone array for sound localisation, speech recognition, and TTS output.
• Vision — Face detection, gaze estimation, emotion recognition (FER), hand-wave detection.
• LLM Integration — RAG-augmented responses via GPT.

Demos

(GIF previews of eye tracking, blinking, mouth articulation, and emotion response coming soon!)

• Eye, eyelid & eyebrow subsystem — asymmetric eyeball motion with cameras, blinking, and brow movement.
• Mouth subsystem — lip articulation for speech and emotion.
• Neck subsystem — full pan and tilt.

Hardware Stack

• ESP32-CAM vision nodes.
• Arduino Mega servo controller.
• Custom servo PCB.
• Embedded microphones.
• Dual-camera eye subsystem.

Subsystems

Mechanical

• Eye, eyelid & eyebrow subsystem — asymmetric eyeball motion with cameras, blinking, and brow movement.
• Mouth subsystem — lip articulation for speech and emotion.
• Neck subsystem — full pan and tilt.

Software

• software/vision/ — face detection, gaze estimation (t_gaze.py), emotion recognition (m_model.py, m_expression.py), hand-wave detection.
• software/speech/ — speech-to-text, TTS, voice selection.
• software/llm/ — GPT interface and RAG pipeline for context-aware responses.
• software/core/ — main driver (vulcan.py, vulcan2.py) coordinating all subsystems.

Electronics

• Servo PCB (ESP32 & Mega variants) — custom KiCad boards in electronics/pcb/.
• ESP32-CAM firmware — electronics/firmware/camera_esp/.
• Arduino servo firmware — electronics/firmware/serial_arduino/.

Repository Structure

project-vulcan/
├── docs/                   # Install guides (CUDA, OpenCV, dlib), diagrams
├── software/
│   ├── core/               # Main entry points
│   ├── vision/             # CV modules
│   ├── speech/             # STT / TTS
│   ├── llm/                # GPT + RAG pipeline
│   │   └── rag/
│   ├── data/               # Knowledge base files for RAG
│   ├── utils/              # Shared helpers
│   ├── tools/              # Diagnostic scripts (check mic, list cameras)
│   ├── experiments/        # Prototypes and WIP scripts
│   ├── models/             # Model weights (gitignored — see models/README.md)
│   ├── archive/            # Deprecated code
│   └── requirements.txt
├── electronics/
│   ├── pcb/                # KiCad projects
│   ├── firmware/           # Arduino / ESP32 sketches
│   └── scripts/            # Python hardware-communication scripts
└── mechanical/             # CAD files (SolidWorks, Fusion, Blender, STL)
    ├── Eye_subsystem/
    ├── Mouth_Subsystem/
    ├── Head Structure/
    ├── Neck Base/
    └── base_holder/

Model weights are not tracked in git — see software/models/README.md for download links.
Install guides for CUDA, OpenCV, and dlib are in docs/

Getting Started

# Install Python dependencies
pip install -r software/requirements.txt

# Download model weights (see software/models/README.md), then run
python software/core/vulcan2.py

Contributor Guidelines

• New features should be isolated into modules.
• Avoid editing archived or legacy systems.
• Experimental work belongs in software/experiments/.
• Shared utilities belong in software/utils/.
• All hardware communication layers must remain abstracted.

Roadmap

• ROS2 migration.
• Improved facial animation pipeline.
• Real-time multimodal emotion fusion.
• Autonomous interaction routines.
• Enhanced gaze tracking.
• Full conversational memory system.

Built With

• Python
• OpenCV
• dlib
• LangChain / OpenAI
• KiCad
• ROS

Downloading all the dependencies for running auto folder

• Make sure cmake and cuda are already downloaded before doing this
• Download the reurequirements.txt file
• pip install -r requirements.txt

Limitations

• The robot covers only the structure above the neck — no full-body mobility.

Core Contributors

• Parth Shah
• Ritwik Sharma
• [Aryan Goyal]
• [Kevin B. Matthew]
• Indrajit Mandal
• Anirudh Singh Air
• [Ayush R. Srivastava]
• Aditya Kushwaha
• Vedant Vakharia
• [Rajas Kasar]