This paper presents an exact quality of service (QoS)-based radar resource management (RRM) method that manages multiple radar sensors, while also taking task dependencies into account. This contrasts with other approaches that mostly do not consider dependencies and assume independence among the tasks. However, this simplification is generally incorrect. Our method optimizes the tasks' assignment to the radar sensors as well as the selection of the radar control parameters jointly, while also exploiting dependencies between tasks. The problem is formulated and solved as mixed-integer linear programming (MILP) optimization problem. A simulated tracking scenario shows that by exploiting task dependencies in combination with coordinating the task assignments among the sensors, our RRM method can improve the operational performance significantly in comparison to the benchmark algorithms.

This paper extends the single radar resource management approach Q-RAM for multiple radar sensors. It optimizes sensor task assignments as well as their radar control parameters jointly based on quality of service. Our method uses single sensor performance models and hence avoids the large parameter space of multi-sensor performance models, which increases exponentially with the number of sensors. This reduces the computational requirements significantly as well as the possible reaction time. The trade-off to this reduction is that our algorithm does not support the execution of the same task on multiple sensors during the same planning period. Though, over multiple planning periods, the tracker can still receive measurements originating from different sensors which allows to exploit the spatial geometry of widely separated sensors. Hence, our method is especially well suited for applications with restricted processing power that need fast reaction times. A simulated tracking scenario shows that the proposed method achieves significantly superior performance to a locally optimized resource management approach as well as to a distance-based assignment. It further accomplished almost identical performance to an approach that employs an exhaustive search to find the best sensor task assignment.

The defence against unmanned aerial vehicles (UAVs) has become an essential topic in recent years. A possible solution that works as an effector against enemy UAVs employs a swarm of its own UAVs. Such a scenario can be modelled as a pursuit evasion scenario, which has been considered in the literature before. A possible solution uses a reinforcement learning approach in which a neural network steers the UAVs. However, previous approaches using multi layer perceptrons (MLPs) have an important caveat that their input dimension is fixed. This severely limits the flexibility of this approach, as changing the number of units in a swarm would require a model to be retrained. This paper presents a solution that employs a Deep Sets based model, allowing the user to change the number of agents inside a swarm as desired. It is shown that using Deep Sets is a viable method to solve a pursuit evasion scenario, in which the number of agents can vary between scenarios, but the trained model stays the same.

This paper addresses the problem of localizing a stationary RF emitter with a mobile UAV, equipped with a single directional antenna. By rotating around its vertical axis, it determines a bearing towards the emitter. Our interest is in optimally selecting the measurement positions to achieve a fast localization. The majority of such systems described in the literature use greedy planning to select the next measurement position. This work experimentally tests an algorithm that performs a non-myopic planning until the final localization step. The algorithm is based on the policy rollout principle and showed good performance in previous simulative studies. It is adapted to match the needs of a real world setup and evaluated in flight trials. Adaptions include the avoidance of close range measurements to prevent inaccurate measurements at high elevation, and the filtering of poor measurements.

The performance of data fusion and tracking algorithms often depends on parameters that not only describe the sensor system, but can also be task-specific. While for the sensor system tuning these variables is time-consuming and mostly requires expert knowledge, intrinsic parameters of targets under track can even be completely unobservable until the system is deployed. With state-of-the-art sensor systems growing more and more complex, the number of parameters naturally increases, necessitating the automatic optimization of the model variables. In this paper, the parameters of an interacting multiple model (IMM) filter are optimized solely using measurements, thus without necessity for any ground-truth data. The resulting method is evaluated through an ablation study on simulated data, where the trained model manages to match the performance of a filter parametrized with ground-truth values.

Reinforcement learning (RL) is already widely applied to applications such as robotics, but it is only sparsely used in sensor management. In this paper, we apply the popular Proximal Policy Optimization (PPO) approach to a multi-agent UAV tracking scenario. While recorded data of real scenarios can accurately reflect the real world, the required amount of data is not always available. Simulation data, however, is typically cheap to generate, but the utilized target behavior is often naive and only vaguely represents the real world. In this paper, we utilize multi-agent RL to jointly generate protagonistic and antagonistic policies and overcome the data generation problem, as the policies are generated on-the-fly and adapt continuously. This way, we are able to clearly outperform baseline methods and robustly generate competitive policies. In addition, we investigate explainable artificial intelligence (XAI) by interpreting feature saliency and generating an easy-to-read decision tree as a simplified policy.

Policy rollout is a method for the online computation of future costs in approximate dynamic programming and has been utilized for various problems, including sensor management. In previous work, it has predominately been applied to the selection of actions from discrete sets. In this article, we present methods for action selection from continuous sets and analyze their tradeoffs. The methods are evaluated on the problem of sensor path planning, with the intent of minimizing the time to localize an emitter using bearing measurements.

This paper considers the problem of finding the best action in a policy rollout algorithm. Policy rollout is an online computation method used in approximate dynamic programming. We applied two different versions of the knowledge gradient (KG) policy to a sensor path planning problem. The goal of this problem is to localize an emitter using only bearing measurements. To the authors' knowledge, this was the first time the KG was applied in a policy rollout context. The performance of the KG policy was found to be comparable with methods used in prior work while also having a potentially wider applicability.

This article describes how resource management techniques developed over the last decades exploit perception-action cycles to improve multifunction radar performance. It is then shown how these existing techniques can be extended by increasing the exploitation of knowledge, information, and understanding. Finally, the potentially drawbacks of cognitive radar techniques are discussed.

Reinforcement learning is the problem of autonomously learning a policy guided only by a reward function. We evaluate the performance of the Proximal Policy Optimization (PPO) reinforcement learning algorithm on a sensor management task and study the influence of several design choices about the network structure and reward function. The chosen sensor management task is optimizing the sensor path to speed up the localization of an emitter using only bearing measurements. Furthermore, we discuss generic advantages and challenges when using reinforcement learning for sensor management.

This paper explores the problem of localizing an emitter with a mobile sensor platform using noisy bearing measurements. It is assumed, that the measuring procedure requires the platform to be stationary for a certain amount of time. Therefore, there exists a trade-off between using time to take one or more measurements and moving to a more advantageous position for observing a target. Using a rollout based algorithm we optimize the time necessary until a given localization accuracy is reached and compare the performance of this algorithm with several algorithms found in literature.

Using recent advances exploiting compressed sensing (CS) theory for radio frequency emitter localization a simple one antenna direction finder has been constructed in order to develop and evaluate theses methods. A series of experiments has been conducted whose results are presented in this paper. Furthermore, the parametrization of the algorithm is analyzed and evaluated. The experiments confirm that CS techniques enhance the accuracy of bearing measurements compared to other methods while at the same time the overall hardware system can be kept simple.

This paper considers the problem of trajectory optimization for two platforms which perform bearings only tracking for multiple targets. It is not assumed that the measurements contain identifying features that enable measurements to be associated to targets. Therefore, ghost targets can appear, which strongly affect the resulting error. In comparison to existing approaches, we propose a trajectory optimization method that considers both the ambiguity in the data association as well as the track error. It is shown that this approach can resolve ghost targets much quicker, resulting in a significantly lower estimation error.

This paper presents an approach for detection and tracking a micro-UAV using the multistatic radar NetRAD. Experimental trials were performed using NetRAD allowing for analysis of real data to assess the difficulty of detection and tracking of a micro-UAV target. The UAV detection is based on both time domain and micro-Doppler signatures, in order to enhance the discrimination between ground clutter and UAV returns. This micro-Doppler based procedure is shown to improve the clutter/target discrimination, in comparison to a Doppler-shift based procedure. The tracking approach is able to compensate for the limited quality measurement generated by each bistatic pair by fusing the measurements available from multiple bistatic pairs.

Previous works have identified key characteristics of a cognitive radar, such as knowledge exploitation, perception, action, memory, intelligence and attention. In this work, it is argued that the cognitive characteristic of anticipation can also enhance radar performance. In this paper it is shown that radar management using a partially observable Markov decision process (POMDP) enables the radar to act with anticipation. A method using policy rollout is applied to approximate a POMDP for a target tracking control problem. Through a simulated example it is demonstrated how the anticipative method departs from a purely adaptive approach, and the subsequent improvement in performance is quantified.

A performance model for the radar search function is derived, which can be used for resources management in a multifunction radar that uses an electronically steered array antenna. The proposed resource allocation model incorporates scenario information, such as the platform altitude and velocity, as well as the areas where targets are expected to pop-up. The model is demonstrated by simulation in airborne and ground based radar scenarios, where a significant improvement in performance is shown in comparison to a non-optimised, uniform resource allocation. This improvement equates to an enlargement of the surveillance region.