Toward Autonomous Robotic Inspections Of Nuclear Facilities Using Directionally-sensitive Neutron Detectors

Whether safeguarding Iranian uranium enrichment facilities, denuclearizing North Korea, or verifying limits on U.S. and Russian arsenals, nuclear safeguards and arms-control traditionally require intrusive on-site inspections to perform verification tasks. In such applications, the ability to localize a radioactive source is imperative for identifying anomalies when no significant neutron emitters are expected or declared to be present, such as for confirming the absence of clandestine withdrawal stations in the centrifuge hall of a gas-centrifuge enrichment plant or undeclared warheads in a storage facility. We investigate the role of autonomous mobile robots, which, if designed properly, may be more effective and efficient and less intrusive than their human counterparts. Towards developing such a capability, we have constructed an “inspector bot,” comprised of three boron-coated straw detectors azimuthally distributed within a cylinder of high-density polyethylene, which is mounted on an omni-directional robotic platform. While many reported methods for source localization use only total detected counts, our inspector bot is specifically designed to provide directional and spectral sensitivity, in addition to gross counts, by utilizing the signals from the three detectors. The detection system has been extensively characterized by MCNP modeling, which has been benchmarked to experiments conducted at the Princeton Plasma Physics Laboratory. For source localization using our inspector bot, we utilize a simplified system of equations which, with the three detectors, is solved online to estimate the direction to the source. We apply the result of the online model in the framework of a “particle filter.” Each “particle” represents a hypothesized source intensity and location, and is assigned a weight according to a cumulative likelihood function. By leveraging the directional sensitivity of our inspector bot, we correctly estimate the location and intensity of a source with fewer measurements, as compared to particle filtering based only on gross counts. Our efforts in advancing this methodology are ongoing as we consider the effects of background noise and obstacles.