M.Sc.A. student · Research collaborator

Pablo BoitelReinforcement learning and control for physical systems

I study reinforcement learning and control for physical systems, especially how methods trained in simulation behave when dynamics and operating conditions change.

ÉTS Montréal × Mila

Dive

01 · Current research

When the model changes, what still works?

I study learning-based control under changing dynamics and imperfect models. My current application is HVAC control, while the broader questions concern robustness, transfer, and physical systems.

R / 01

Learning across models

Training across varied models to reduce dependence on any one simulation.

R / 02

Reinforcement learning and control

Studying how policies and controllers behave when dynamics and conditions change.

R / 03

Physical systems

Applying these ideas to engineering systems, currently HVAC.

02 · Publication

Published work

Open water · − 50–180 m

2026
Pablo Boitel · Kun Zhang

Generalizing HVAC Control With Domain Randomized Reinforcement Learning

A study of how domain randomization can help an HVAC controller transfer across different building models.

View publication

ACM Sustainability Week '26
pp. 383–386

03 · CV / − 180–400 m

Research and engineering experience

My path into research has moved from physical modeling and control to numerical simulation and reinforcement learning.

Academic CVPDF · Synced from GitHub

04 · About / − 400–800 m

Below the work

I came to research through mathematics, mechanical systems, and control. I like moving between theory, simulation, and hands-on work.

I have also taught mathematics and worked in a FabLab. Outside research, I sail, cycle, play badminton, fly gliders when I can, and go to art-house cinemas.