Colin Bellinger

Colin Bellinger is a computer scientist whose reinforcement-learning research spans costly measurements, robotic learning, and incremental policy updates. He earns computer science degrees at Carleton and uOttawa, then researches machine learning at the Alberta Machine Intelligence Institute, Dalhousie University, and the National Research Council of Canada. His work includes agents that choose when to measure scientific systems, a deep reinforcement-learning report on an industrial robotic arm, and Action Value Gradient, tested with a manipulator and a mobile robot. Since 2025, he is an associate professor at the University of Ottawa and a Vector Institute faculty affiliate, leading the Machine Intelligence and Robot Learning group.

Now at: School of Electrical Engineering and Computer Science and Vector Institute
Badges: Wayfarer

Timeline

2007–2016

Computer science degrees

He earns a bachelor's degree in computer science with a mathematics minor at Carleton in 2007, a Master of Computer Science there in 2010, and a PhD in computer science at uOttawa in 2016.
May 2016–July 2018

Postdoctoral fellow at Amii

From May 2016 to July 2018, he works as a postdoctoral fellow at the University of Alberta with the Alberta Machine Intelligence Institute (Amii), researching machine learning.
July 2018–October 2018

Donald Hill Fellow at Dalhousie

From July 2018 to October 2018, he is a Donald Hill Fellow in Computer Science at Dalhousie University, where he researches machine learning.
October 2018–August 2025

Staff scientist at NRC

He joins the National Research Council of Canada as a staff scientist in October 2018 and works on machine learning and reinforcement learning through August 2025.
2021

Choosing when to measure

In 2021, he coauthors a framework that lets reinforcement-learning agents select actions and decide whether to measure a system's state at each step. Dueling DQN and PPO agents learn action policies with up to 50% fewer state measurements.
2023

Deep RL on a UR10e arm

In 2023, he coauthors a technical report on visual tracking and reaching with deep reinforcement learning on an industrial UR10e robotic arm. The report's initial experiments find PPO learns a more stable policy with less data than deep Q-learning.
2024

Action Value Gradient

In 2024, he coauthors Action Value Gradient (AVG), an incremental deep policy-gradient method that learns from recent samples without batch updates or replay buffers. The work demonstrates deep reinforcement learning with a robotic manipulator and a mobile robot using incremental updates.
Since 2025

Associate professor at uOttawa

Since 2025, he is an associate professor at the University of Ottawa's School of Electrical Engineering and Computer Science and a faculty affiliate at the Vector Institute. He leads the Machine Intelligence and Robot Learning (MIRL) group and develops reinforcement-learning methods for robotic manipulation and navigation, materials discovery, and health and safety decisions.