CMU Researcher Uses Deep Reinforcement Learning to help Control Nuclear Fusion Reactions

    The process of joining two hydrogen nuclei to create a single, heavier nucleus is known as nuclear fusion. Massive amounts of energy are released throughout this process. The fundamental issue, however, continues to be maintaining the levels required for adding electricity to the grid. Nuclear fusion, which happens naturally at the sun’s center when extremely high temperatures and pressures are present, is the only process that can cause hydrogen nuclei to fusion. Physicists have also accomplished nuclear fusion in thermonuclear weapons, although these cannot be used as energy sources. 

    Utilizing magnetic fields to keep a hydrogen plasma at the proper temperature and pressure to fuse the nuclei’s outer shells is another way to create nuclear fusion. This procedure occurs inside a tokamak, a huge apparatus that confines hydrogen plasma in a torus shape using magnetic fields. It takes countless micromanipulations to the magnetic fields and bursts of extra hydrogen particles to keep the plasma contained and keep its shape. Few large-scale tokamaks are currently in operation that can support this kind of research. Hence time to do tests on them is highly sought after. The only facility running in the United States is the DIII-D National Fusion Facility.

    The first reinforcement learning system to regulate the magnetic field enclosing the fusion reaction was developed by DeepMind, a division of Google’s parent company, Alphabet, that specializes in artificial intelligence. The plasma was successfully held stable in the lab and was molded into various shapes. DeepMind conducted its investigation in Lausanne, Switzerland, on the Variable Configuration Tokamak (TCV) and reported its findings in Nature. However, a student from the School of Computer Science (SCS) at Carnegie Mellon University just made a significant advancement. This represents a massive step toward harnessing the enormous power generated by nuclear fusion as a source of cheap, abundant, clean energy because he was able to apply reinforcement learning to manage nuclear fusion reactions.

    The Ph.D. candidate from the machine learning department employed reinforcement learning to manage the hydrogen plasma of the tokamak machine at the DIII-D National Fusion Facility in San Diego. He was the first CMU researcher to experiment on the desired machines, the first to employ reinforcement learning to modify the rotation of a tokamak plasma, and the first to apply reinforcement learning on the largest machine in operation in the country. The Princeton Plasma Physics Laboratory (PPPL) was also involved in the study. Reinforcement learning leverages information from prior attempts to provide the best results. RL algorithms were employed in this experiment to analyze historical and current data to control and adjust the speed of the plasma’s spin in search of the best stability. It was found that reinforcement learning impacted the plasma’s pressure and spin, which was a significant first step in creating clean energy. 

    When more hydrogen atoms are fired into the plasma, it rotates. Potentially, stabilizing the plasma and making it simpler to contain might be accomplished by varying the speed of these fired particles. We used two learning algorithms in this test. The first entailed teaching how the plasma behaves using tokamak data gathered over several years. The second algorithm assesses the plasma’s state and determines how quickly and in what direction to inject extra particles to change the plasma’s speed. The research’s immediate objective is to provide physicists with the means to induce this differential rotation so they may conduct tests that could increase plasma stability. The long-term objectives include employing reinforcement learning to regulate other plasma state components and eventually reach pressures and temperatures long enough to support a power plant. This would open the door for everyone to access endless, clean energy.

    The doctoral researcher showed that algorithms could regulate the rotational speed of the plasma. This was the first instance of rotation control using reinforcement learning. The experiment still requires some further testing because it is not yet flawless. However, the researcher won high appreciation for how he led an experiment at a national fusion laboratory and applied the theoretical information he had obtained at CMU to a real fusion challenge.

    This Article is written as a research summary article by Marktechpost Staff based on the research paper 'Magnetic control of tokamak plasmas through deep reinforcement learning'. All Credit For This Research Goes To Researchers on This Project. Check out the paper and reference article.
    Please Don't Forget To Join Our ML Subreddit

    Asif Razzaq is an AI Journalist and Cofounder of Marktechpost, LLC. He is a visionary, entrepreneur and engineer who aspires to use the power of Artificial Intelligence for good.

    Asif’s latest venture is the development of an Artificial Intelligence Media Platform (Marktechpost) that will revolutionize how people can find relevant news related to Artificial Intelligence, Data Science and Machine Learning.

    Asif was featured by Onalytica in it’s ‘Who’s Who in AI? (Influential Voices & Brands)’ as one of the ‘Influential Journalists in AI’ ( His interview was also featured by Onalytica (

    Recent Articles


    Related Stories

    Stay on op - Ge the daily news in your inbox