How it started
How it started
VRAI (Vanier Research in Artificial Intelligence) started as a student-run research lab at Vanier College. Founding members Thavin Thanabalasingam and Le Tuan Huy (Tony) Nguyen with some help from the Math Faculty dedicated hours of work to embodied AI, teaching physical robots to perceive, reason, and act in the real world.
What they worked on
Their early effort focused on bringing together imitation learning, computer vision, and edge computing into a single cohesive system. Using Hugging Face's LeRobot as their primary framework, they taught physical robots by recording human demonstrations—allowing the robot to observe, learn the underlying behaviors, and replicate actions autonomously.
To perceive the physical environment, the robots relied on multi-camera arrays whose visual streams were processed through convolutional networks to provide a rich spatial understanding. All inference was engineered to run locally on NVIDIA Jetson hardware without any cloud dependency, making the platforms fully self-contained so the robot could process information and make decisions directly onboard.
Their pipeline
The development workflow operated as a continuous end-to-end pipeline. It began by assembling the robot, flashing the compute environments, and meticulously calibrating every motor joint and camera feed until teleoperation was responsive and precise. From there, data collection took place: a human operator controlled a leader arm while the paired follower arm mirrored the motion in real time, capturing joint positions and multi-angle camera feeds into high-quality demonstration datasets.
They then fed these demonstrations into imitation learning architectures—primarily Action Chunking Transformers (ACT)—training the policies until the robot could reliably replicate the demonstrated skills. Finally, the optimized model was deployed onboard the Jetson, allowing the robot to execute tasks autonomously using only its joint sensors and real-time vision.
Evolution & impact
What started as an ambitious student initiative has since gained momentum. Following a recruitment, the team widened significantly as new students across diverse science and engineering programs at Vanier College joined the lab. The project's academic and practical promise was recognized with competitive funding, securing a $1,000 InGenious grant as well as a $5,000 FRQNT internship grant under the supervision of Math Faculty mentor Florian Duquerroix.
Today, the project remains very active. The prototype built by the founding team now serves as an active, hands-on platform where new Vanier students are learning artificial intelligence and modern Machine Learning.
$1,000
Student innovation grant awarded to support the lab
$5,000
Supervised research internship under Florian Duquerroix
Still Active
New Vanier students learning modern ML on the prototype
Hardware
- Hugging Face LeRobot SO-101 dual-arm setup (leader + follower)
- NVIDIA Jetson (Jetpack 6.2) for onboard inference and teleoperation
- Multi-camera arrays (side view + wrist/grip view)
- Yahboom ROSMASTER X3 Plus (archived)
Frameworks & tools
- Hugging Face LeRobot — dataset collection, training, and inference
- ACT (Action Chunking Transformer) — their primary imitation learning model
- ROS — used during the Yahboom phase for sensor integration
- PyTorch — model training and experimentation
Curious how the model actually works?
Read more about the architecture they used.