Radar-based human activity recognition (RadHAR) is an attractive alternative to wearables and cameras because it preserves privacy, is contactless, and is robust to occlusions. However, dominant convolutional neural network (CNN)- and recurrent neural network (RNN)-based solutions are computationally intensive at deployment, and recent lightweight vision transformer (ViT) and state-space model (SSM) variants still exhibit substantial complexity. In this article, we present RadMamba, a parameter-efficient, micro-Doppler-oriented Mamba SSM tailored to radar HAR under on-sensor compute, latency, and energy constraints typical of distributed radar systems. RadMamba combines 1) channel fusion with downsampling; 2) Doppler-aligned segmentation that preserves the physical continuity of Doppler over time; and 3) convolutional token projections that better capture Doppler-span variations, thereby retaining temporal–Doppler structure while reducing the number of Floating-point Operations/Inference ( $\#$ FLOP/Inf.). Evaluated across three datasets with different radars and types of activities, RadMamba matches the prior best 99.8% accuracy of a recent SSM-based model on the continuous wave (CW) radar dataset, while requiring only $1/400$ of its parameters. On a dataset of non-continuous activities with frequency-modulated continuous wave (FMCW) radar, RadMamba remains competitive with leading 92.0% results using about $1/10$ of the parameters, and on a continuous FMCW radar dataset it surpasses methods with far more parameters by at least 3%, using only $6.7{\,}\mathrm {k}$ parameters. Code is available at https://github.com/lab-emi/AIRHAR

This roadmap provides a comprehensive and forward-looking perspective on the individualized application and safety of non-ionizing radiation (NIR) dosimetry in diagnostic and therapeutic medicine. Covering a wide range of frequencies, i.e. from low-frequency to terahertz, this document provides an overview of the current state of the art and anticipates future research needs in selected key topics of NIR-based medical applications. It also emphasizes the importance of personalized dosimetry, rigorous safety evaluation, and interdisciplinary collaboration to ensure safe and effective integration of NIR technologies in modern therapy and diagnosis.

To characterize atmospheric turbulence, the Doppler moments are estimated by weather radars. However, moment accuracy is highly sensitive to radar transmission parameters such as pulse repetition time (T_s) and number of pulses (N_p), which affect Doppler ambiguity and estimation variance. Traditional fixed-parameter radars face trade-offs between aliasing and measurement precision. This paper proposes an adaptive radar framework that dynamically adjusts T_s and N_p on a per-scan basis to improve Doppler moment estimation at a single resolution cell level. Inspired by the Fully Adaptive Radar (FAR) concept, the method also includes a novel multi-lag Doppler width estimation scheme. Results demonstrate enhanced estimation accuracy, enabling better responsiveness to localized and non-stationary weather conditions.

We consider the problem of the large amount of data produced by current radar systems, particularly Frequency Modulated Continuous Wave (FMCW) radars, with the goal of minimizing latency and memory usage in signal processing pipelines for edge-based applications. Drawing inspiration from event-based cameras, we propose a novel event-based radar perception method that utilizes only two single snapshots to achieve performance levels suitable for current radar applications. After mathematically deriving the approach, we validate its effectiveness through both simulations and real-world experiments involving multiple targets.

This article presents the first subject-specific head pose estimation approach using only one frequency-modulated continuous wave radar data frame. Specifically, the proposed method incorporates a deep learning framework to estimate head pose rotation and orientation frame-by-frame by combining a convolutional neural network operating on range-angle radar plots and a PeakConv network. The proposed method is validated with an in-house collected dataset, including annotated head movements that varied in roll, pitch, and yaw, and these were recorded in two different indoor environments. It is shown that the proposed model can estimate head poses with a relatively small error of approximately 6.7°–14.4° for all rotational axes and is capable of generalizing to unseen, new environments when trained in one scenario (e.g., lab) and tested in another (e.g., office), including in the cabin of a car.

The problem of radar-based multi-target tracking for indoor human monitoring is considered. Tracking and counting the number of people moving as a group is particularly challenging as multiple individuals are close together and their radar signatures are mixed. A transformer-based classification approach for counting the number of grouped people is proposed. The Neural Network model is trained with selected features from the spatial domain and Doppler frequency domain, which are concatenated over multiple frames to form a sequence for the transformer network. Compared to statistical classifiers, the self-attention mechanism allows transformers to capture feature long-term dependencies. The proposed classifier is integrated into a tracking pipeline in order to monitor the position and number of people in the grouped targets. The method proposed is experimentally verified using a 24 GHz Frequency Modulated Continuous Wave (FMCW) radar with 250MHz bandwidth. Despite the relatively coarse range resolution, the proposed method achieves 92.5% accuracy in these initial tests. Furthermore, the method performances and related accuracy is analyzed according to various parameters.

Classification of human activities performed sequentially and with unconstrained durations using radar sensors has been studied in this work. A novel processing pipeline comprising a sequence segmentation stage, a segment processing stage, and a classification stage has been proposed to address this challenge. Specifically, the segmentation stage has been implemented by monitoring Rényi entropy for fluctuations in the radar data, with the entropy, derived from micro-Doppler spectrograms, functioning as a descriptive quantity of the activity being performed. The method has been experimentally verified on a challenging, publicly available dataset collected with a network of five simultaneously operating pulsed ultrawideband radars. Classification performance has been compared to reference works in the literature on the same dataset, and a test accuracy and macro F1-score of 89.3% and 82.0% have been, respectively, demonstrated.

The role of radar for building situation awareness in (semi)autonomous vehicles is severely restricted by its low angular resolution. The physical size of the radar, which determines its antenna aperture size and thus the radar angular resolution, is often a subject of stringent limitations to physically fit the system in the vehicles. Multiple input multiple output systems are used to increase the achievable angular resolution, and these are often combined in the literature with algorithms inspired by synthetic aperture radar techniques that exploit the velocity of the vehicle itself for finer resolution. Some of the most common approaches are reviewed, in this context, with a specific focus on challenges for the implementation of data collected in real driving scenarios. Key experimental results using representative algorithms and driving data collected in the city of Delft, The Netherlands, are presented and discussed.

The problem of reconstructing 3D signatures of human activities for monitoring and classification is considered in this work. A method based on data fusion from distributed MIMO (multiple-input multiple-output) radar nodes is developed in order to generate 3D intensity maps and related voxel-wise velocity vectors. The proposed method was evaluated with a dataset collected using three 60 GHz radars and including 7 activities performed by 30 participants. The results show that both static postures and dynamic activities can be captured effectively: consecutive phases of activities/movements can be identified by combining spatial intensity and velocity vectors, and the participant can be localized in the area under test. Furthermore, initial promising classification results of 98.3% macro F1-score are demonstrated in a three-class problem using the proposed 3D intensity maps and velocity vectors as inputs to a Convolutional Neural Network classifier.

The problem of estimating breathing rates and detecting apnea events with radars located at an elevated and tilted position is considered in this paper. This is particularly relevant in psychiatric clinics, where radars (or other sensors) must be installed out of reach of patients. In this work, a feasibility study is presented, using an experimental setup with two 60 GHz Frequency-Modulated Continuous Wave (FMCW) radars placed at 2.7 m height at a tilted angle towards the participants, and one radar at 1 m height looking straight to the participants, who are sitting and lying on the floor. A new comprehensive dataset with 30 participants and 7 activities was collected with this setup. Using phase extraction and filtering, the work presents an apnea detection probability of up to 90 % with an elevated radar, and comparable mean error rates for breathing estimation of below 2 respirations per minute (rpm) for all radars. The results show that the respiration data and apnea detection from all radar positions are comparable. This proves the feasibility of the proposed radar deployment positions, benefiting application fields such as psychiatric care.

Monitoring gait symmetry reliably is crucial, as it is an early indicator of Parkinson's disease (PD). In this work, a method is presented to analyze gait asymmetries using a 24-GHz frequency-modulated continuous-wave (FMCW) multiple-input-multiple-output (MIMO) radar in a nonclinical environment. The proposed method is validated by analyzing the gait of 60 people recorded in an environment that presents multiple challenges such as multipath, gait interruption events, and different trajectories, which cause aspect angle variability. This article presents algorithms to address these challenges to extract useful gait information from the radar data and discusses the performance of the proposed system, analyzing the ratio of correct feet identifications and the accuracy of the asymmetry gait parameters. It achieved a mean absolute error of the symmetry ratio below 8% for all gait parameters. This shows that the proposed system and algorithms are robust for both clinical and in-home implementations.

This paper presents a Joint Deep Probabilistic Subsampling framework with Cramér-Rao Bound (CRB) Integration (J-DPSC), which optimizes both transmit and receive antenna selection independently, enabling robust DoA estimation in mono-static and bi-static radar modes. Our method incorporates a differentiable worst-case CRB computation for end-to-end optimization. Additionally, sidelobe level (SLL) suppression is naturally integrated into the training process through data constraints, without requiring explicit penalty terms. Compared to prior methods, J-DPSC significantly reduces model complexity and the number of trainable parameters while maintaining high DoA estimation accuracy. Beampattern results demonstrate the effectiveness of our approach in improving performance with an acceptable SLL, making it well-suited for practical sparse array design in automotive radar applications.

Traditional target tracking using monostatic radar systems typically rely on centralized or decentralized architectures, where all data is transmitted to a fusion center for estimating the position and velocity of mobile agents. This approach introduces a single point of failure and can significantly increase communication costs, particularly when the fusion center is far from individual radar nodes. To overcome these issues, we introduce a distributed Alternating Direction Method of Multipliers (ADMM) for target localization using a radar network, wherein each radar node shares its observed data only with its immediate neighboring nodes, and achieves consensus with the radar network on the estimated target locations and velocities. We perform simulations incorporating critical system parameters such as the number of radar nodes and Signal-to-Noise Ratio (SNR) to assess their impact of estimation accuracy and convergence speed of the proposed distributed ADMM algorithm. We highlight the additional benefits of our proposed solution, and present directions for future work.

The problem of enabling adaptive capabilities in the context of weather radar is considered in this paper. Inspired by the cognitive radar framework, an approach based on Reinforcement Learning (RL) is formulated to deal with the monitoring of multiple storm cells moving near a potential area of interest. The approach aims to dynamically adjust the radar waveform bandwidth, and consequently maximum measurable range and range resolution, in order to provide the best monitoring based on a purposely-defined reward function. The approach is successfully validated with a simulator developed in Python & StoneSoup. Results demonstrate that the proposed method outperforms traditional fixed-scan ('sit and spin') strategies commonly used in weather radar operations.

The problem of estimating instantaneous distributed target velocity using noisy measurements by multiple asynchronous automotive radar sensors is investigated. Two novel neural networks (NNs)-based approaches are proposed to address the problem. Both NNs use the point cloud with radar detections as an input. In the first approach, a hybrid NN is designed to take a set of points inside a cluster as its input, extract spatial-dynamic features to be used as weights for each input point, and apply them to obtain a weighted least square (WLS) solution for instantaneous velocity estimation. To this end, dedicated loss functions are proposed to allow the model to predict weights that can follow a velocity profile curve satisfying target constraints. Moreover, a small offset in the radial velocity value of each point is applied to adjust errors in the sensor measurements. In the second approach, a deep NN is proposed that takes a set of points inside a cluster as its input and directly outputs velocity estimates. Both approaches have been verified experimentally using the large open-source automotive RadarScenes dataset. The results show a significant improvement in terms of mean absolute error in velocity estimation over the state-of-the-art alternative techniques. Moreover, the estimated velocity is used as an additional measurement value inside a target tracker. Results show that this can increase the performance of the tracker, especially during challenging scenarios such as abrupt changes in the velocity of the target.

The problem of radar-based tracking of groups of people moving together and counting their numbers in indoor environments is considered here. A novel processing pipeline to track groups of people moving together and count their numbers is proposed and validated. The pipeline is specifically designed to deal with frequent changes of direction and stop-and-go movements typical of indoor activities. The proposed approach combines a tracker with a classifier to count the number of grouped people; this uses both spatial features extracted from range-azimuth (RA) maps and Doppler frequency features extracted with wavelet decomposition. Thus, the pipeline outputs over time both the location and the number of people present. The proposed approach is verified with experimental data collected with a 24-GHz frequency-modulated continuous-wave (FMCW) radar. It is shown that the proposed method achieves 93.15% accuracy in terms of counting the number of people and a tracking metric optimal subpattern assignment (OSPA) of 0.335. Furthermore, the performance is analyzed as a function of different relevant variables such as feature combinations and scenarios.

The problem of diminished unambiguous target velocity interval induced by the time-division-multiplex mode (TDM) of multiple-input-multiple-output (MIMO) frequency-modulated continuous-wave (FMCW) automotive radar has been explored. A novel MIMO antenna array activation mode and a parametric approach for Doppler de-aliasing based on a two-step cross-entropy optimization are proposed. The TDM Doppler signal model has been derived, and a novel two-step cost function is proposed to achieve robust and efficient estimation. In contrast to the state-of-the-art method, the method proposed does not need multiple overlapped antennas and can resolve multiple targets in the same range and Doppler bins. The proposed method has been verified with numerical simulations with different parameter settings, as well as experimental data from a radar target simulator.

In this paper, an automatic labelling process is presented for automotive datasets, leveraging on complementary information from LiDAR and camera. The generated labels are then used as ground truth with the corresponding 4D radar data as inputs to a proposed semantic segmentation network, to associate a class label to each spatial voxel. Promising results are shown by applying both approaches to the publicly shared RaDelft dataset, with the proposed network achieving over 65% of the LiDAR detection performance, improving 13.2% in vehicle detection probability, and reducing 0.54 m in terms of Chamfer distance, compared to variants inspired from the literature.

In this article, the classification of dynamic vulnerable road users (VRUs) using polarimetric automotive radar is considered. To this end, a signal processing pipeline for polarimetric automotive MIMO radar is proposed, including a method to enhance angular resolution by combining data from all polarimetric channels. The proposed signal processing pipeline is applied to measurement data of three different types of VRUs and a car, collected with a custom automotive polarimetric radar, developed in collaboration with Huber+Suhner AG. Several polarimetric features are estimated from the range-velocity signatures of the measured targets and are subsequently analyzed. A Bayesian classifier and a convolutional neural network (CNN) using these estimated polarimetric features are proposed and their performance is compared against their single-polarized counterparts. It is found that for the Bayesian classifier, a significant increase in classification performance is achieved, compared to the same classifier using single polarized information. For the CNN-based classifier, utilizing the distribution of polarimetric features of the target’s range-velocity signatures also increases classification performance, compared to its single-polarized version. This shows that polarimetric information is valuable for classification of VRUs and objects of interest in automotive radar.

The problem of radar-based, continuous Human Activity Recognition (HAR) has been studied in this work. A fixed-window segmentation method based on dual timescales has been proposed to tackle this challenge. The method is experimentally validated on a challenging publicly available dataset with 14 participants and 9 activities, and is compared to reference works from the literature. L1PO validation of the method yields a test accuracy and macro F1-score of 87.5 % and 80.1 % respectively.