Offline RL trains tokamak fusion reactor control using historical data

Researchers have developed an offline reinforcement learning approach to control the plasma rotation profile within tokamaks, devices crucial for fusion energy research. This method trains solely on historical data from the DIII-D tokamak, addressing the challenge of controlling complex, high-dimensional dynamics without accurate simulators. The developed policy, which utilizes probabilistic models for training rollouts, has shown promising results when deployed on the DIII-D Tokamak. AI

IMPACT Demonstrates a new method for applying RL to control complex physical systems using historical data, potentially applicable to other scientific and engineering domains.

RANK_REASON Academic paper detailing a novel application of reinforcement learning to a complex scientific problem. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• DIII-D

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →