Smart Radiation Sensor Management Radiation Search and Mapping using Mobile Robots

Abstract

MOTIVATION The current geopolitical situation requires automated tools for quick and effective assessement of threats. Modern threats are subtle and ephemeral, and can be hidden across large areas. Classical information extraction methods, where data is randomly collected and then subsequently filtered and analyzed by human operators in search of particular signatures, are no longer effective against today’s modern threats. Data collection must be guided by querying world models that afford the span and resolution needed for multi-scale problems. Currently, searching for radiation sources is usually done manually, by operators waving radiation counters in front of them as they walk. This method does not provide any visual or statistical data map of the area in question. To quickly characterize the severity of the situation, an efficient way of obtaining this radiation map is needed. When searching for a weak radiation source, a speck of uranium for example, manual methods are unlikely to yield results. In nuclear search, the strength of the signal relative to noise (SNR) falls with the square of the distance R to the source, as the latter increases. The relation between SNR and distance motivates bringing the sensor as close to the source as possible [1]. Mobile robots can carry sensors close to the source, and position them accurately for required measurement collection. Using traditional sequential testing theory we can only confirm the presence of a source of a particular strength at a given location. For locations where these specific nuclear signatures are not detected, no information is given regarding the local radiation levels. A different approach is therefore needed if the objective is to map the radiation intensity over a certain area. In this article we suggest two different motion planning strategies for radiation map building. The first, named the gradient-based Bayesian method, an uncertainty metric is used to define a potential function with which to bias the search towards particular areas of the map where uncertainty regarding radiation levels is highest. The second strategy, named the sequential-based Bayesian method, the robot visits every area cell along a pre-determined path, and the time it spends at each cell depends on the local uncertainty levels. The sequential-based Bayesian method ensures that each cell is only visited once, and thus it is time-optimal. However, due to the motion plan of the sensor being predetermined, parts of the map that could be potentially the most interesting could be revealed last. In addition, this method is not suitable when the prior is time-varying, that is, in the case of dynamic environments, since areas explored once are not revisited. The gradient-based Bayesian method, on the other hand, offers an approximate map of varying confidence at every time step, but it requires longer time for the completion of the map. However, the method outperforms the sequentialbased Bayesian mapping in the initial stages of the area scanning, suggesting that when time constraints are imposed that will not allow the sequential-based method to terminate, a better map can be obtained with the gradient-based method. In addition, the gradient-based Bayesian method can accomodate real-time changes in the environment, through an on-line adaptation of the function that generates potential field. The methods described in this article are not only suited to applications of “nuclear forensics,” where we need to determine in the least possible time, and at a given probability of a false positive, whether fissile material has been processed in a given area; they are also applicable to the problem of assessing the contamination due to accidental or malicious release of radioactive isotopes. As a result of a radiation map, decision makers can single out safe from unsafe regions and quantify contamination as a first step towards containment and cleanup.