Sridhar Mahadevan

About

Sridhar Mahadevan is a prominent researcher whose work has fundamentally shaped the fields of reinforcement learning (RL) and autonomous robotics. Best known for his landmark 1996 paper on average reward reinforcement learning — which has garnered over 400 citations — Mahadevan challenged the conventional discounted reward paradigm, providing rigorous foundations and algorithms that better capture long-term agent behavior in continuing tasks. His early contributions to robot learning, including pioneering work on applying RL to behavior-based robots using the subsumption architecture, helped establish the viability of machine learning for real-world robotic systems at a time when the field was still nascent. Mahadevan's research evolved to tackle the challenge of scaling RL to complex, multi-agent environments through hierarchical frameworks. His development of Cooperative Multi-agent Hierarchical RL and investigations into Hierarchical Partially Observable Markov Decision Processes (HPOMDPs) for robot navigation demonstrated how structured decomposition can dramatically accelerate learning. With contributions spanning theoretical foundations, algorithm design, and practical robotics applications — collectively amassing hundreds of citations — Mahadevan's work remains essential reading for researchers seeking to bridge the gap between abstract RL theory and deployable intelligent systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Average reward reinforcement learning: Foundations, algorithms, and empirical results

   401 citations · 1996
2. 2
   Automatic programming of behavior-based robots using reinforcement learning

   152 citations · 1992
3. 3
   Hierarchical multi-agent reinforcement learning

   129 citations · 2006
4. 4
   Robot Learning

   108 citations · 1993
5. 5
   Scaling Reinforcement Learning to Robotics by Exploiting the Subsumption Architecture

   65 citations · 1991
6. 6
   Hierarchical Multiagent Reinforcement Learning

   56 citations · 2004
7. 7
   Rapid Task Learning for Real Robots

   53 citations · 1993
8. 8
   Learning Hierarchical Partially Observable Markov Decision Process Models for Robot Navigation

   50 citations · 2001
9. 9
   Approximate planning with hierarchical partially observable Markov decision process models for robot navigation

   45 citations · 2003
10. 10
    To Discount or not to Discount in Reinforcement Learning: A Case Study Comparing R Learning and Q Learning

    37 citations · 1994

Key Collaborators

Contact & Links