Student | AI | Robotics | RMIT University
Developing a hybrid system that allows the Tiago robot to dynamically adjust task execution based on natural language instructions, combining PDDL planning, ROSPlan, and low-level behavior trees.
As part of a group project, I worked on developing an autonomous indoor delivery robot using a ROSBot platform. The robot uses computer vision to detect ArUco markers, which act as pickup and delivery location indicators, and navigates through indoor spaces without relying on GPS.
My main contribution to the project was building the ArUco marker detection and localisation system. I implemented real-time marker detection using OpenCV, and designed a stable pose estimation pipeline that integrates with ROS2’s TF2 framework for accurate global localisation. To make the system reliable, I addressed practical challenges such as time synchronisation issues, unstable Z-axis pose estimates, and TF transform errors, all of which are common in real-world robotics applications.
I also contributed to setting up Nav2, the robot’s autonomous navigation stack. This included helping integrate the vision system with navigation and visual servoing, so the robot could approach the correct location and align precisely for pickup or delivery.
This project gave me hands-on experience with real-time perception, localisation, and autonomous navigation in ROS2, while also teaching me the importance of debugging and refining robotics systems under practical conditions.
Key Areas I Worked On: • ArUco Marker Detection and Pose Estimation • TF2-based Localisation Pipeline • Real-Time ROS2 Integration • Nav2 Setup and Visual Servoing Integration • Debugging Transform and Time Synchronisation Challenges
Role: Hazard Detection Developer (Group Project)
Framework: ROS2 Autonomous Robot System
Contributed to the development of an autonomous robot capable of exploring unknown environments, detecting hazards, and returning to the start position.
Tools & Technologies: ROS2, Python, TF2, RViz, Computer Vision
As part of a biomedical machine learning project, I developed a deep learning system for classifying colon cell histopathology images into cancerous vs. non-cancerous and cell type categories (fibroblast, inflammatory, epithelial, others). The project used a modified version of the CRCHistoPhenotypes dataset containing 27x27 RGB images from 99 patients.
I designed and trained convolutional neural networks (CNNs) with techniques such as L2 regularization, dropout, batch normalization, and focal loss to improve model robustness and handle class imbalance. The final model achieved 90% accuracy and 0.85 F1-score for cancer detection, and 81% accuracy and 0.67 macro F1-score for cell type classification.
Additionally, I explored semi-supervised learning by incorporating unlabeled data, providing practical insights into its benefits and limitations for real-world medical AI applications.
Project Type: AI Planning & Optimization
Tools Used: Answer Set Programming (ASP), Clingo, Python
This project focused on generating fair and feasible match schedules for the LZV Cup, a non-professional indoor football league. The tournament follows a time-relaxed round-robin format with real-world constraints like team availability, rest periods, and fairness requirements.
We modelled the scheduling problem using Answer Set Programming (ASP) — a declarative logic-based approach well-suited for complex constraint satisfaction tasks. The Clingo solver was used to compute valid schedules under these constraints.
To handle instance preparation, result processing, and validation, we developed a supporting Python pipeline.
This project demonstrated how ASP, when combined with well-designed heuristics, can rival traditional optimization methods like Tabu Search. The experience deepened my understanding of constraint modeling, solver tuning, and hybrid AI system design.
Vehicle Electronics | Quick Shifting | BSPD | Performance Optimization
I led the Electronics Team for our Formula 6 car, competing in the Formula Bharat student racing competition. My work focused on designing and developing reliable, efficient, and competition-compliant electronic systems to enhance the vehicle’s performance and safety.
Key Contributions:
Technologies & Tools:
Embedded Systems, Circuit Design, MATLAB, Vehicle Safety Systems, Performance Tuning
This project allowed me to apply practical electronics design and problem-solving skills in a competitive motorsport environment.