Projects |

Autonomous Tunnel Burrowing Robot

Thu, 02 Jul 2026 00:00:00 +0000

Overview

The Autonomous Tunnel Burrowing Robot is an intelligent robotic excavation platform designed to autonomously construct underground tunnels based on user-defined dimensions. The system combines a mobile robot, a gantry-mounted robotic arm, environmental perception, and autonomous planning to perform excavation tasks in challenging environments.

This project was developed during my research at the University of Notre Dame under the supervision of Dr. Yasemin Ozkan Aydin.

Motivation

Underground construction is often dangerous, labor-intensive, and difficult to automate. This project explores how autonomous robotics can improve safety, precision, and efficiency in tunnel excavation by integrating perception, manipulation, and autonomous decision-making.

Key Features

Autonomous tunnel excavation
Mobile robotic platform
Gantry-mounted robotic manipulator
User-defined tunnel generation
ROS and Gazebo simulation
Sim-to-real deployment
Soil interaction analysis
Autonomous task execution

Hardware

Custom mobile robot
Gantry robotic arm
Intel RealSense Depth Camera
Torque sensor
Custom excavation end-effectors

Software

ROS
Gazebo
Python
C++
OpenCV
Point Cloud Processing
Robot Manipulation

Research Contributions

This project investigates:

Autonomous excavation planning
Robot manipulation for construction tasks
Perception using depth cameras
Soil interaction analysis
Sim-to-real robotic deployment
Autonomous navigation during excavation

Experimental Evaluation

The system was evaluated using multiple excavation scenarios with different soil conditions and interchangeable digging tools. Both simulation and real-world experiments demonstrated reliable autonomous excavation and tunnel generation.

Future Work

Future research directions include:

Learning-based excavation strategies
Vision-language robot control
Multi-robot cooperative excavation
Foundation models for autonomous construction robotics

Project Status

Status: ✅ Completed Research Project

OK-1 Quadruped Robot

Thu, 02 Jul 2026 00:00:00 +0000

Overview

OK-1 is a fully custom-designed quadruped robot developed as an end-to-end robotics platform, covering mechanical design, electronics, embedded systems, software development, simulation, and autonomous control. The project was created as part of OK Dynamics, a robotics startup that I co-founded, with the goal of building an affordable and flexible research platform for legged robotics.

Motivation

Commercial quadruped robots are often expensive and difficult to customize. OK-1 was developed to provide an accessible platform for robotics research, rapid prototyping, and experimentation with autonomous locomotion, perception, and artificial intelligence.

Key Features

Fully custom quadruped robot
Complete mechanical design
Embedded control system
ROS and Gazebo simulation
Computer vision integration
Autonomous locomotion
Wireless remote control
Modular hardware architecture

Mechanical Design

The entire robot was designed from scratch using SolidWorks, including:

Chassis
Legs
Servo mounting system
Electronics enclosure
Remote controller
Complete CAD assembly

The design emphasizes modularity, lightweight construction, and ease of maintenance.

Electronics

Custom power distribution
Servo motor control
Embedded microcontroller
Raspberry Pi integration
Wireless communication
Battery management

Software Stack

C++
Python
ROS
Gazebo
OpenCV
TensorFlow
Embedded C

Research Contributions

The OK-1 platform served as a foundation for exploring:

Quadruped locomotion
Inverse kinematics
Gait generation
Robot simulation
Autonomous navigation
Vision-based perception
AI-enabled robotics

Experimental Results

The platform was successfully deployed in both simulation and real-world environments. It was used to validate locomotion algorithms, embedded control systems, and perception pipelines while serving as a versatile platform for robotics research and education.

Future Work

Future developments include:

Reinforcement learning for locomotion
Vision-language-action models
Autonomous outdoor navigation
Multi-modal perception
Onboard AI acceleration
Swarm robotic coordination

Project Status

Status: ✅ Completed Research & Startup Project

Salamander Robot

Thu, 02 Jul 2026 00:00:00 +0000

Overview

The Salamander Robot project investigates how spinal joint actuation can improve the locomotion efficiency, stability, and adaptability of bio-inspired quadruped robots. The project combines reinforcement learning with robotic simulation and real-world experiments to study coordinated spine–limb motion across different terrains.

Motivation

Many quadruped robots rely solely on leg motion for locomotion. Inspired by salamanders and other sprawling animals, this research explores how an actively controlled spinal joint can improve mobility and robustness.

Key Features

Bio-inspired quadruped robot
Deep Reinforcement Learning (DQN)
Active spinal joint
ROS and Gazebo simulation
Sim-to-real deployment
Raspberry Pi onboard control
Terrain adaptability analysis

Hardware

Custom Salamander Robot
Dynamixel Servo Motors
Raspberry Pi
Custom mechanical design

Software

ROS
Gazebo
Python
C++
Deep Q-Network (DQN)

Research Contributions

The project investigates:

Coordinated spinal and limb dynamics
Reinforcement learning for locomotion
Terrain adaptation
Efficient gait generation
Bio-inspired robot control

Publications

Related publications include work presented at robotics workshops and conferences on bio-inspired locomotion and reinforcement learning.

Future Work

Future research includes:

Vision-based locomotion
Model-based reinforcement learning
Outdoor terrain adaptation
Foundation models for robot locomotion

Project Status

Status: ✅ Completed Research Project

This project was developed during my research at the University of Notre Dame under the supervision of Dr. Yasemin Ozkan Aydin.

TinyDrone-MicroNav

Thu, 02 Jul 2026 00:00:00 +0000

Overview

TinyDrone-MicroNav is an autonomous navigation framework for nano aerial vehicles that enables safe and efficient navigation using onboard perception and lightweight planning algorithms. The project focuses on resource-constrained platforms such as the Crazyflie 2.1 and aims to bridge the gap between simulation and real-world deployment.

Motivation

Autonomous navigation remains challenging for ultra-lightweight drones because of their limited sensing, computation, memory, and battery capacity. This project investigates how lightweight perception, planning, and learning algorithms can enable reliable navigation in dynamic environments.

Key Features

Vision-based autonomous navigation
Dynamic obstacle avoidance
Local path planning
Global recovery strategy
Webots simulation
Crazyflie 2.1 deployment
Simulation-to-real transfer
Lightweight algorithms for embedded platforms

Hardware

Crazyflie 2.1
Multi-ranger Deck
Flow Deck
Onboard IMU

Software

Python
Webots
Crazyflie Python Library
OpenCV
NumPy

Current Research

Current work focuses on integrating:

Deep Reinforcement Learning
Vision-Language-Action Models
TinyML
Embedded AI
Efficient onboard perception

to improve autonomous navigation under severe computational constraints.

Future Work

Future directions include:

End-to-end visual navigation
Multi-drone collaboration
Real-world outdoor navigation
Embedded deployment of foundation models
Fully onboard autonomous flight

Project Status

Status: 🚧 Active Research Project

This project is currently under active development as part of my PhD research at the University of Illinois Chicago.

Vision-Language-Action Navigation

Thu, 02 Jul 2026 00:00:00 +0000

Overview

Vision-Language-Action (VLA) Navigation is an ongoing research project that aims to develop a universal autonomous navigation framework capable of controlling both aerial and ground robots directly from visual observations and natural language instructions.

The goal is to eliminate traditional modular navigation pipelines by learning an end-to-end policy that maps perception and language directly to robot actions.

Research Motivation

Current robotic navigation systems typically consist of separate perception, localization, mapping, planning, and control modules. While effective, these pipelines are often difficult to generalize across different robot platforms and environments.

Inspired by recent advances in foundation models and embodied AI, this project investigates whether a single Vision-Language-Action model can generalize navigation behaviors across multiple robotic platforms.

Research Objectives

Universal navigation across multiple robot platforms
End-to-end visual navigation
Natural language robot control
Cross-domain generalization
Sim-to-real deployment
Resource-efficient inference

Target Platforms

Autonomous Tiny Drones
Mobile Robots
Autonomous Vehicles
Legged Robots

Input Modalities

RGB Images
Video Streams
Natural Language Instructions
Robot State Information

Example instructions include:

“Navigate to the red chair.”
“Avoid pedestrians.”
“Move to the charging station.”
“Fly through the open doorway.”

Output

The model directly predicts navigation actions such as:

Linear velocity
Angular velocity
Steering commands
Drone velocity commands
Waypoints

Planned Architecture

The proposed framework combines:

Vision Encoder
Language Encoder
Multimodal Fusion
Action Decoder

to produce end-to-end navigation commands.

Simulation Platforms

NVIDIA Isaac Sim
NVIDIA Omniverse
CARLA
Webots
ROS 2

Potential Applications

Autonomous driving
Drone navigation
Warehouse automation
Search and rescue
Industrial inspection
Service robotics

Current Status

This project is currently under active development as part of my PhD research at the University of Illinois Chicago.

Future Directions

Future work includes:

Multi-robot collaboration
Long-horizon navigation
Outdoor deployment
Continual learning
Foundation models for autonomous robotics
Deployment on embedded robotic platforms

PyTorch

Thu, 26 Oct 2023 00:00:00 +0000

PyTorch is a Python package that provides tensor computation (like NumPy) with strong GPU acceleration.