Saminda Abeyruwan

About

Saminda Abeyruwan is a robotics and machine learning researcher whose work sits at the intersection of reinforcement learning, sim-to-real transfer, and high-speed robotic control. He is perhaps best known for his pioneering contributions to robotic table tennis, a demanding testbed that requires millisecond-precision perception, planning, and actuation. His papers on learning high-speed precision table tennis on physical robots (2022–2023, accumulating over 30 citations combined) demonstrated that reinforcement learning systems could achieve sustained multi-shot rallies with human players and return balls to precise targets — milestones long considered out of reach for learned policies. His i-Sim2Real framework advanced sim-to-real transfer by incorporating tight human-robot interaction loops during training, addressing a critical gap in prior approaches. More recently, his work culminated in a landmark achievement: the first learned robot agent to reach amateur human-level performance in competitive table tennis. He has also contributed to humanoid robot motion synthesis, wearable activity recognition, and the Gemini Robotics initiative, which seeks to bring large multimodal AI models into physical robotic systems. Across his career, Abeyruwan has consistently pushed the boundaries of what autonomous robotic agents can accomplish in dynamic, real-world environments.

Research Focus

Key Achievements

Top Papers

1. 1
   Robotic Table Tennis: A Case Study into a High Speed Learning System

   17 citations · 2023
2. 2
   Learning High Speed Precision Table Tennis on a Physical Robot

   16 citations · 2022
3. 3
   i-Sim2Real: Reinforcement Learning of Robotic Policies in Tight Human-Robot Interaction Loops

   12 citations · 2022
4. 4
   Motion Capture and Contemporary Optimization Algorithms for Robust and Stable Motions on Simulated Biped Robots

   5 citations · 2013
5. 5
   Single- and Multi-channel Whistle Recognition with NAO Robots

   4 citations · 2015
6. 6
   Activity monitoring and prediction for humans and NAO humanoid robots using wearable sensors

   4 citations · 2015
7. 7
   Gemini Robotics: Bringing AI into the Physical World

   4 citations · 2025
8. 8
   Achieving Human Level Competitive Robot Table Tennis

   3 citations · 2025
9. 9
   Achieving Human Level Competitive Robot Table Tennis

   3 citations · 2024
10. 10
    RLLib: C++ Library to Predict, Control, and Represent Learnable Knowledge Using On/Off Policy Reinforcement Learning

    2 citations · 2015

Key Collaborators

Contact & Links