Julien Mairal

About

Julien Mairal is a researcher whose work sits at the intersection of robotic learning, reinforcement learning, and human-robot knowledge transfer. His research focuses on developing methods that enable robots to acquire complex manipulation skills through observation and demonstration, tackling fundamental challenges in bridging the gap between human and robotic action spaces. Among his most notable contributions, Mairal has explored how human demonstration videos can serve as scalable, cost-effective training signals for robotic systems. His 2023 work on learning reward functions by observing humans (16 citations) addresses the difficult problem of transferring manipulation skills across fundamentally different observation and action spaces — a key bottleneck in practical robot learning. Complementing this, his 2021 research on residual reinforcement learning from demonstrations (11 citations) extended the residual RL framework to handle visual inputs and sparse rewards, making learning from demonstrations more tractable in realistic settings. Mairal's contributions are particularly significant for advancing sample-efficient robot learning, reducing reliance on costly engineered reward functions, and making robotic systems more adaptable through human-guided supervision — work that holds strong promise for real-world deployment of intelligent robotic manipulators.

Research Focus

Key Achievements

Top Papers

1. 1
   Learning Reward Functions for Robotic Manipulation by Observing Humans

   16 citations · 2023
2. 2
   Residual Reinforcement Learning from Demonstrations

   11 citations · 2021

Key Collaborators

Contact & Links