- Research Article
- 10.1115/1.4070433
Automated Robotic Exploration of Noisy Radiation Fields from Raw Real-World Radiation Data using a Dynamically Enhanced Implementation of Bayesian Optimization
- Nov 20, 2025
- Journal of Nuclear Engineering and Radiation Science
- Jeremy Marquardt + 4 more +4
Abstract Assisted automation of radiation field survey with robots has the potential to increase the efficiency of maintenance in radiation environments, reduce human exposure to harmful radiation fields, accelerate emergency response, and improve anomaly characterization. Algorithms for the control and path planning of these robots need to be accurate, efficient, and interpretable, so human operators can make informed decisions to update the search path on-the-fly. Several approaches have been used to address this problem with Bayesian optimization recently showing promising results. Bayesian optimization has only been demonstrated in relatively simple simulated radiation environments using static measurements which minimize the amount of information collected. In this paper, we expand Bayesian optimization by accounting for dynamic real-time data acquisition to optimize the agents' collection routes. We demonstrate the proposed approach using noisy raw datasets obtained from a real-world radiation environment in three different scenarios combined with a probabilistic model. The radiation data were collected in different room-sized environments, including a research reactor at power with attenuation obstacles, and at significantly different gamma radiation levels. The performance of the dynamic collection scheme is compared with traditional point to point static data collection. It is shown that the addition of dynamic measurements can effectively find at least 10-36% hotter signals and reduce the mean number of loop iterations required to survey the space by up to 70%, depending on scenario. Radiation field reconstruction accuracy is improved considerably, with a 40% reduction in root mean squared error in most cases.
Read more