Vitalis Vosylius
I am a PhD student at Imperial College London, supervised by Edward Johns at The Robot Learning Lab. My research interests
include robot manipulation and the broader field of robot learning.
Before joining Imperial for my MSc in AI and subsequently my PhD, I studied Applied Physics
and worked on advancing laser beam engineering techniques at the Center for Physical Sciences and Technology.
Recently, I also spent time as a research intern at the Dyson Robot Learning Lab, led by Stephen James.
If you would like to get in touch to chat or collaborate with me, feel free to send me an
e-mail!
Email  / 
Google Scholar
 / 
LinkedIn
|
|
Research
My research interests lie in the areas of robot learning and manipulation. Specifically, I
focus on algorithmically improving sample efficiency and generalisation capabilities of
imitation learning frameworks. This allows robots to acquire useful manipulation skills more
effectively from limited data and adapt to new tasks and settings quickly.
Below are some of my projects that illustrate my efforts in this area.
|
|
Render and Diffuse: Aligning Image and Action Spaces for Diffusion-based
Behaviour Cloning
Vitalis Vosylius,
Younggyo Seo,
Jafar Uruç,
Stephen James,
RSS, 2024
project page
We introduce R&D, a method that integrates RGB observations with low-level actions through
3D renders of the robot and iteratively updates them using a learnt denoising process,
significantly improving learning efficiency and spatial generalisation.
|
|
Few-Shot In-Context Imitation Learning via Implicit Graph Alignment
Vitalis Vosylius,
Edward Johns,
CoRL, 2023
project page
We learn how to align graph representations of objects and use it as a foundation of a
few-shot in-context imitation learning framework.
|
|
DALL-E-Bot: Introducing Web-Scale Diffusion Models to Robotics
Vitalis Vosylius,
Ivan Kapelyukh,
Edward Johns,
RA-Letters, 2023
project page
We use Web-Scale Image Diffusion Models to generate goal images for object rearrangement
tasks.
|
|
Where To Start? Transferring Simple Skills to Complex Environments
Vitalis Vosylius,
Edward Johns,
CoRL, 2022
project page
We use graph-based affordance model to find suitable starting configurations for executing
different tasks in complex environments.
|
|