Miguel Abreu

About

Miguel Abreu is a robotics and artificial intelligence researcher whose work sits at the intersection of deep reinforcement learning and humanoid robot locomotion, with a particular focus on autonomous robotic soccer. Through his contributions to the RoboCup 3D Simulation League — most notably as part of the FC Portugal team, which claimed the 2023 World Championship title — Abreu has helped push the boundaries of what autonomous agents can learn without hand-crafted domain knowledge. His most influential work explores how reinforcement learning algorithms, particularly Proximal Policy Optimization, can train humanoid robots to master low-level motor skills entirely from scratch, including running, kicking while in motion, and maintaining balance under perturbation. These papers have collectively garnered over 100 citations, reflecting strong community interest. Beyond locomotion primitives, Abreu has tackled higher-level soccer strategy learning and developed hybrid frameworks that blend model-based analytical controllers with learned residual policies for greater robustness. More recently, his research has addressed nuanced challenges such as imperfect biological symmetry in policy learning, a novel direction with real implications for natural robot movement. His body of work offers students a compelling roadmap for applying modern machine learning to complex, real-world robotic control problems.

Research Focus

Key Achievements

Top Papers

1. 1
   Learning to Run Faster in a Humanoid Robot Soccer Environment Through Reinforcement Learning

   37 citations · 2019
2. 2
   Learning low level skills from scratch for humanoid robot soccer using deep reinforcement learning

   31 citations · 2019
3. 3
   Humanoid Robot Kick in Motion Ability for Playing Robotic Soccer

   20 citations · 2020
4. 4
   6D Localization and Kicking for Humanoid Robotic Soccer

   11 citations · 2021
5. 5
   Learning high-level robotic soccer strategies from scratch through reinforcement learning

   6 citations · 2019
6. 6
   Addressing imperfect symmetry: A novel symmetry-learning actor-critic extension

   4 citations · 2024
7. 7
   Learning hybrid locomotion skills—Learn to exploit residual actions and modulate model-based gait control

   4 citations · 2023
8. 8
   Designing a skilled soccer team for RoboCup: exploring skill-set-primitives through reinforcement learning

   3 citations · 2025
9. 9
   Robust biped locomotion using deep reinforcement learning on top of an analytical control approach

   2 citations · 2021
10. 10
    FC Portugal: RoboCup 2023 3D Simulation League Champions

    2 citations · 2024

Key Collaborators

Contact & Links