A joint research team from UNIST and Korea University has developed a technology that utilizes quantum computing to more efficiently train artificial intelligence (AI) that is resilient to environmental changes. They explained that by resolving the computational bottleneck problem in reinforcement learning, the potential for application in fields such as robotics and autonomous driving has been enhanced.

UNIST announced on the 6th that a research team led by Professor Sung-Hwan Yoon and a research team led by Professor Joong-Heon Kim at Korea University have developed 'QRIM (Quantum Robust Inner Minimization),' a robust reinforcement learning technique based on quantum algorithms. The research results have been accepted for the International Conference on Machine Learning (ICML 2026).

Reinforcement learning is a method in which AI learns optimal behavioral strategies through trial and error. However, it has a limitation in that performance degrades significantly if the actual environment differs from the learning environment.

Robust reinforcement learning, developed to complement this, proceeds with learning by assuming a worst-case scenario, but it has the problem of significantly increased computational cost because it must repeatedly explore possible environmental changes.

The research team solved this problem by utilizing the superposition properties of quantum computing. QRIM combines Quantum Amplitude Estimation (QAE) and Quantum Minimum Search (QMF) algorithms to more efficiently identify the most unfavorable situation among multiple environment candidates.

According to the research team, while existing methods had to check all N environments, QRIM can achieve the same results with a √N level of search.

The research team proved through theoretical analysis that QRIM provides a computational acceleration effect at the square root level compared to existing methods.

In addition, it was stated that robustness was achieved in FrozenLake and CartPole environment experiments using only about 20–30% of the computational load compared to existing methods. The number of environment searches required for learning was found to have decreased by about 64–80% compared to existing methods.

In addition, the research team explained that the technology operated normally in experiments using IBM's 127-qubit quantum computer and maintained performance even in environments with quantum hardware noise.

Professor Yoon Seong-hwan stated, “This is an example showing that quantum computing can complement the limitations of existing artificial intelligence,” adding, “It could be utilized in fields such as autonomous driving, robotics, and medicine, where responding to environmental changes is important.”