Papers
35
Total Citations
456
H-Index
13
About
Abbas Abdolmaleki is a prominent researcher at the intersection of reinforcement learning (RL) and robotics, with a particular focus on making RL algorithms practical, robust, and deployable in real-world settings. His most influential contributions center on offline reinforcement learning, where his "Keep Doing What Worked" framework — cited over 100 times across its publications — introduced behavioral modeling priors to constrain off-policy learning from fixed datasets, a critical advance for applying RL to physical systems like robot control. Abdolmaleki has also made significant strides in robust RL, developing frameworks that handle model misspecification and sim-to-real transfer challenges that frequently undermine real-world deployment. His work on constrained and multi-objective policy optimization addresses the practical need for safe, smooth control signals in continuous robotic tasks, moving beyond naive bang-bang solutions. From humanoid soccer locomotion to complex robotic stacking of diverse geometric shapes, his research portfolio spans both foundational algorithmic development and compelling applied demonstrations. With a growing citation record across multiple high-impact publications and contributions spanning hybrid discrete-continuous action spaces, Abdolmaleki has established himself as a versatile and rigorous contributor to the advancement of intelligent, deployable robotic systems.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4Value constrained model-free continuous control30 citations · 2019
- 5Continuous-Discrete Reinforcement Learning for Hybrid Control in Robotics27 citations · 2020
- 6A Distributional View on Multi-Objective Policy Optimization23 citations · 2020
- 7Learning a Humanoid Kick with Controlled Distance20 citations · 2017
- 8Omnidirectional Walking and Active Balance for Soccer Humanoid Robot17 citations · 2013
- 9
- 10Beyond Pick-and-Place: Tackling Robotic Stacking of Diverse Shapes16 citations · 2021