Undergraduate Honours Projects

I supervise undergraduate honours projects in machine learning, reinforcement learning, and robotics. These projects are designed to provide students with hands-on experience in modern AI research, including problem formulation, implementation, evaluation, and scientific communication.

Students will gain experience with PyTorch, reinforcement learning frameworks, robotics platforms, and real-world experimental design.

Featured Projects

Featured Collaboration

Privacy-Preserving Synthetic Health Data Generation
Ongoing Undergraduate Research

This project focuses on developing generative models for producing high-quality synthetic health data while preserving patient privacy. The work explores trade-offs between data utility and privacy guarantees, with applications in healthcare analytics and machine learning.

Collaboration: Vector Institute, TELUS Health

Featured Collaboration

Sim-to-Online Reinforcement Learning for Robotics Control
Ongoing Research Collaboration

This project investigates sim-to-online reinforcement learning for robotic control. An initial control policy is learned efficiently in a JAX-based simulation using Soft Actor-Critic (SAC), and then An initial control policy is learned efficiently in a JAX-based simulation using Soft Actor-Critic (SAC), and then transferred to a physical robotic system. to a physical robotic system. The policy is further adapted online using incremental deep reinforcement learning approaches inspired by recent work, the Action Value Gradient (AVG) methods. The aim is faciliate efficient, adaptable and robustness embodied agents.

Collaboration: Amii, National Research Council (NRC)

Related Work

Past Student Projects

Reinforcement Learning with Pre-Collected Datasets
Fall 2025

Tara Denaud Joseph, Richa Kewalramani, Fay Lee

Reinforcement Learning and Its Applications on the Kinova Gen3 Lite Robotic Arm
Fall 2025

Zahra Suleymanova, Vishal Bhat

Multi-Agent Reinforcement Learning for Safe Hovering and Navigation on Crazyflie Drones
Fall 2025

Michael O’Sullivan, Aydin Yalcinkaya

Single-Agent Reinforcement Learning for Drone Robotics
Fall 2025

Kevin Naveen, Chad B. Yassin, Pronoy Fuad

Is PPO Inherently Safe?
Winter 2026

Estelle Ngounou, Eliel Beonao, Jean-Philippe Nahimana Bahenda

Available Project Areas

Reinforcement Learning (RL)

Single-Agent Reinforcement Learning
Develop and evaluate RL algorithms in simulated environments. Projects may explore:

Sample efficiency and exploration strategies
Safe and constrained decision-making
Goal-conditioned learning
Learning under limited or noisy data

Multi-Agent Reinforcement Learning
Study how multiple agents learn to cooperate or compete. Possible directions:

Coordination and communication between agents
Decentralized vs. centralized training
Emergent behaviors in complex environments

Robotics and Sim-to-Real Learning

Sim-to-Real Transfer with Crazyflie Drones
Develop RL or control methods in simulation and transfer them to real Crazyflie quadrotors. Topics include:

Domain randomization
Robust control policies
Perception-based control

Robotic Manipulation with Kinova Arms
Design learning-based approaches for robotic manipulation tasks:

Grasping and object interaction
Vision-based control
Policy learning from limited data

Mobile Robotics with TurtleBots
Develop navigation and decision-making algorithms for ground robots:

Autonomous navigation and planning
Sensor-aware decision-making
Reinforcement learning for robotics

Data-Efficient and Adaptive Learning

Fine-Grained Active Measurement in RL
Investigate how agents can selectively acquire observations when sensing is costly:

Observation-cost-sensitive decision making
Adaptive data collection strategies
Exploration under sensing constraints

Continual Reinforcement Learning
Design agents that can learn across multiple tasks without forgetting:

Lifelong learning
Transfer across environments
Avoiding catastrophic forgetting

Machine Learning for Challenging Data Settings

Learning from Imbalanced Data
Develop deep learning methods that perform well when data is skewed or rare:

Class imbalance and rare event detection
Synthetic data generation
Robust evaluation methods

Privacy-Preserving Health Data Generation
Design generative models for sensitive datasets:

Privacy-preserving synthetic data
Evaluation of utility vs. privacy trade-offs
Applications in healthcare datasets

Expectations

Students undertaking honours projects are expected to:

Work independently with regular supervision
Read and understand research papers
Implement and evaluate machine learning models
Present results in written and oral form

Strong programming skills (Python) and familiarity with machine learning are expected. Prior experience with deep learning or reinforcement learning is an asset but not required.

How to Apply

If you are interested in an honours project, please email me.

Please include:

A transcript (unofficial is fine)
A brief statement of interest
Any relevant experience (courses, projects, GitHub)

Colin Bellinger