Takashi Onishi

About

Takashi Onishi is a researcher at the forefront of reinforcement learning (RL) for process control and robust decision-making. His work addresses critical challenges in scaling RL to real-world industrial applications. Onishi's most impactful contribution, "Local Control is All You Need: Decentralizing and Coordinating Reinforcement Learning for Large-Scale Process Control" (2022, 4 citations), pioneers a decentralized RL framework that overcomes the scalability and coordination hurdles of conventional controllers. This approach offers inherent flexibility and generalization to unseen situations, reducing the need for constant parameter tuning. Complementing this, his paper "Learning Robust Options by Conditional Value at Risk Optimization" (2019, 2 citations) introduces a novel method for learning options—temporally extended actions—that are robust to model parameter uncertainty. By optimizing for conditional value at risk, Onishi moves beyond worst-case or average-case thinking, providing a principled way to balance risk and performance in uncertain environments. His work is notable for bridging theoretical RL advances with practical industrial needs, offering a path toward more reliable and autonomous control systems. Onishi's research is essential reading for those interested in scalable, robust RL for complex, real-world systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Local Control is All You Need: Decentralizing and Coordinating Reinforcement Learning for Large-Scale Process Control

   4 citations · 2022
2. 2
   Learning Robust Options by Conditional Value at Risk Optimization

   2 citations · 2019

Key Collaborators

Contact & Links