Hello!

GitHub · LinkedIn · Twitter · E-mail · Scholar

I'm Louis Le Lay, a robotics engineer working on making humanoid robots genuinely agile through reinforcement learning and simulation.

I contributed to the creation of mjlab [ tweet ], an Isaac Lab-style RL framework built on MuJoCo Warp, and I'm an active contributor to Isaac Lab.

At PAL Robotics, I owned the RL locomotion effort for the Kangaroo humanoid end-to-end: built the full pipeline from scratch, led the technical direction, and replaced the MPC-based walk the company had used for over 20 years. The stack is open source (pal_mjlab).

Kangaroo humanoid walking on real hardware with RL locomotion — Real hardware

G1 robot getting up using reinforcement learning — G1 fall recovery

Tennis motion retargeting to a robot — Motion retargeting

Experience

PAL Robotics · Robotics Engineer — Jul 2025 – Mar 2026
University of Tokyo · Research Intern — Jun – Sep 2024
Airbus · Robotics Engineer Intern — Sep 2023 – Jun 2024

Publications

For a complete and up-to-date list of publications, see my Google Scholar profile.

mjlab: A Lightweight Framework for GPU-Accelerated Robot Learning

Kevin Zakka, Qiayuan Liao, Brent Yi, Louis Le Lay, Koushil Sreenath, Pieter Abbeel

arXiv preprint, Jan 2026

paper · website · code

Software

pal_mjlab

PAL Robotics' robots in mjlab. Standard RL interface for Kangaroo, Talos, and TIAGo.

code ·

mjlab

Isaac Lab API on MuJoCo Warp. Led by Kevin Zakka.

code ·

Kinova Gen3 RL & Sim2Real Toolkit

Isaac Lab training to real Kinova Gen3 deployment via ROS 2.

code ·

SO-ARM RL: Manipulation tasks

RL manipulation tasks for SO‑ARM100/101 in Isaac Lab.

code ·

Talks

Training and Deploying RL Agents for Manipulation on the SO-ARM — Talk at ROSCon España, Nov 2025.

Education

Sorbonne University · Master of Engineering, Robotics — 2022 – 2025
ÉTS Montréal · Dual Degree, Information Technology — 2024 – 2025