Dual Radar: A Multi-modal Dataset with Dual 4D Radar for Autononous Driving

AdMaPlace March 18, 2025

0 Comments

Dual Radar: A Multi-modal Dataset with Dual 4D Radar for Autononous Driving

Background & Summary

As the key aspect of autonomous driving technology, environmental perception can timely detect the external things that affect the safety of driving in the process and provide the basis for the subsequent decision-making and control links, which guarantee the safety and intelligence of driving^1,2,3,4,5. In recent years, sensors such as cameras, LiDAR, and radars in ego vehicles have attracted significant research interest due to remarkable increases in the performance of sensors and the computers’ arithmetic power^{6,7,8,9,10,11,12,13,14,15,16,17}.

The camera has high resolution, enabling true RGB information and abundant semantic features such as colors, categories, and shapes. However, when meeting quickly changeable or weak-intensity light, the results obtained from the environmental perception task using only the camera are undesirable^{18,19,20,21,22}. Moreover, monocular cameras cannot accurately acquire distance information, and multi-camera or fisheye cameras suffer from obvious lens distortion problems^23,24. LiDAR collect dense 3D point clouds, with the advantages of high density and accurate precision^25,26. When the autonomous vehicle drives at high speed, The LiDAR continues to operate in a mechanical full-view rotation to collect data, and this driving condition can lead to distortion of the point cloud obtained by the LiDAR²⁷. Meanwhile, LiDAR performs poorly in adverse weather due to its reliance on optical signals to acquire point clouds^28,29,30. Radar, on the other hand, opens up new prospects in autonomous vehicles due to its recognized advantages such as small size, low cost, all-weather operation, high-speed measurement capability, and high range resolution^31,32,33. In particular, radar works with radio wave signals, has good penetration performance in adverse weather, and has a long propagation distance, which makes up for the shortcomings of cameras and LiDAR^34,35,36. Currently, the radar sensors commonly used in autonomous driving include 3D radar and 4D radar. 3D radar provides three-dimensional information, including distance, azimuth, and velocity. Compared to 3D radar, 4D radar adds the object height dimension and can provide denser point cloud data and more azimuth angle information. This gives 4D radar a clear advantage in handling challenging driving conditions^37,38,39.

Due to its sparsity (the density of the electric cloud distribution), 4D radar collects less information than LiDAR point clouds. Although 4D radar performs well in unfavorable scenarios, its sparsity can still lead to the possibility of missing objects. With the development of radar sensors, various types of 4D radar have been applied in scientific research⁴⁰. 4D radar has different working modes according to the detection range, which can be classified into short-range, middle-range, and long-range modes⁴¹. In Table 1, each 4D radar can be observed to have its advantages. In different working modes, the resolution and range of point clouds collected by 4D radars are different, and 4D radars have different point cloud densities and collection ranges in multiple working modes⁴¹. As shown in Fig. 1(d), due to the large beamwidth of Arbe Phoenix, which usually does not process noise, the point cloud is dense, but it will lead to the possibility of false detection. On the other hand, as shown in Fig. 1(e), the point cloud data collected by the ARS548 RDI sensor is relatively sparse. Although it has less noise, this may result in the loss of information about certain objects, negatively impacting the detection of small or nearby objects. However, the denoising process can improve the object detection accuracy of 4D radar. At the same time, the ARS548 RDI can collect a long-range of point clouds than the Arbe Phoenix, which means that the ARS548 RDI can collect information about objects at longer distances. In the field of autonomous driving, existing datasets are usually studied using only one type of 4D radar, lacking comparative analysis of 4D radars with different point cloud densities and levels of noise in the same scenario, as well as a lack of research on perception algorithms that can process different types of 4D radars.

Table 1 Comparison of configurations of 4D radar in commonly datasets.

Full size table

To validate the performance of different types of 4D radar in object detection and object tracking tasks and to fulfill researchers’ needs for a 4D radar dataset, we propose a novel dataset with two types of 4D radar point clouds. Collecting two types of 4D radar point clouds can explore the performance of point clouds with different sparsity levels for object detection in the same scenario, which will provide a basis for developing 4D radar research in the field of ego vehicles. Our dataset has been published and researchers such as Fan⁴², Gao⁴³, and Li⁴⁴ have cited our dataset for comparative analyses with contrasting other 4D radar datasets.

Our main contributions to this work are as follows:

(1) We present a dataset with multi-modal data, which includes camera data, LiDAR point cloud, and two types of 4D radar point cloud. Our dataset can study the performance of different types of 4D radar data, contributes to the study of perception algorithms that can process different types of 4D radar data, and can be used to study single-modal and multi-modal fusion tasks.

(2) Our dataset provides a variety of challenging scenarios, including scenarios with different road conditions (city and tunnel), different weather (sunny, cloudy, and adverse weather), different light intensities (normal light and backlight), different times of day (daytime, dusk, and nighttime), These scenarios can be used to assess the performance of different types of 4D radar point cloud in different scenarios.

(3) Our dataset consists of 151 continuous time sequences, most of which last 20 seconds. There are 10,007 frames carefully synchronized frames and 103,272 high-quality annotated objects.

Methods

In this section, we propose the main aspects of our dataset, including the sensor specification of the ego vehicle system, sensor calibration, dataset annotation, data collection and distribution, and the visualization of the dataset.

Sensor Specification

Our ego vehicle’s configuration and the coordinate relationships between multiple sensors are shown in Fig. 1. The platform of our ego vehicle system consists of a high-resolution camera, a new 80-line LiDAR, and two types of 4D radar. The sensor configurations are shown in Table 2. All sensors have been carefully calibrated. The camera and LiDAR are mounted directly above the ego vehicle, while the 4D radars are installed in front of it. Due to the range of horizontal view limitations of the camera and 4D radars, we collect data only from the front of our ego vehicle for annotation. The ARS548 RDI captures data within an approximately 120° horizontal field of view and 28° vertical field of view in front of the ego vehicle, while the Arbe Phoenix, operating in middle-range mode, collects data within a 100° horizontal field of view and 30° vertical field of view. The LiDAR collects around the ego vehicle in a 360° manner but only retains the data in the approximate 120° field of view in front of it for annotation. The specifications of the sensors are provided in Table 2.

Table 2 The configuration of the autonomous vehicle system platform.

Full size table

Sensor Calibration

The calibration work for this dataset primarily includes camera-LiDAR joint calibration and camera-4D radar joint calibration, both conducted using offline calibration methods. The advantage of offline calibration lies in achieving spatial synchronization in a windless open environment, thereby minimizing spatial synchronization errors and obtaining more precise calibration parameters. Temporal synchronization between different sensors is achieved through hardware synchronization devices, which can provide high-precision timing based on GPS. Additionally, as some scenes in this experiment are derived from environments with weak communication capabilities (e.g., tunnels), offline calibration ensures that temporal synchronization is unaffected by environmental constraints. This allows the synchronization between different sensors to be reliably managed by hardware devices, ensuring the feasibility of the experimental data. Existing 3D radar calibration methods are based on LiDAR calibration techniques and have achieved good results by incorporating the unique characteristics of 3D radar. Therefore, the process of obtaining the internal and external parameters of 4D radar can follow and extend the calibration procedures for 3D radar. In this study, our calibration work is as follows: (1) By placing the calibration board at different positions within the common field of view of the 4D radar and the camera, at least six sets of calibration data are collected. Each set of calibration data includes RGB images from the camera and point cloud data from the 4D radar. One side of the calibration board facing the 4D radar and the camera features a checkerboard pattern, while the center of the other side is equipped with a corner reflector, whose imaging point serves as a feature point. (2) From each set of calibration data, the 3D coordinates of the feature points in the radar coordinate system and the 2D coordinates of the feature points in the camera coordinate system are extracted. (3) Using the correspondence between the 3D coordinates of the feature points in the radar coordinate system and the 2D coordinates in the camera coordinate system, a transformation equation is established. The transformation equation is then solved to obtain the calibration parameters for the 4D radar and the camera. (4) The calibration process for LiDAR and cameras is consistent with that for 4D radar and cameras. Both involve using a specific calibration board or device to collect multiple sets of data within the shared field of view of the two sensors. Feature points are then extracted, and coordinate transformations are performed to establish the spatial transformation relationship between the two sensors, thereby determining the calibration parameters between them. The point cloud projection relationship between the point clouds of the LiDAR and the two radars on the camera image in the same frame is shown in Fig. 2, where the LiDAR point cloud is the complete un-noise-cancelled point cloud data, the point cloud of the ARS548 RDI and the point cloud of the Arbe Phoenix.

Dataset Annotation

In our dataset, we provide 3D bounding boxes, object labels, and tracking IDs for each object. The synchronized frames are annotated based on camera and LiDAR point clouds. The 3D bounding boxes are derived by projecting the LiDAR point cloud onto the camera images. We do not distinguish between objects in dynamic or static states during annotation. To synchronize the timestamps between different sensors, we choose the Precision Time Protocol (PTP) to align the time between multiple sensors using GPS message timing with time synchronization devices.

We provide basic information for each object within a 100-meter distance, including the relative coordinates (x, y, z) of the corresponding 3D bounding box, absolute dimensions (length, width, height), and the orientation angle (alpha) in the Bird’s Eye View (BEV), which requires careful calculation. To ensure the usability of the data, we also provide features for object occlusion and truncation. We annotated over ten categories of labels, with a focus on five primary categories, including “Car”, “Pedestrian”, “Cyclist”, “Bus” and “Truck”. The remaining labels are grouped as “Others”. According to statistics from the collected raw data, approximately 50,000 synchronized frames were extracted, and 10,007 frames were annotated among them. We annotated 103,272 objects from the annotated frames.

Methods for technical validation

We employed a server based on the Ubuntu 18.04 system as our hardware platform. We adopted the OpenPCDet project based on Pytorch 10.2, the batch size setting to the default value of 4, and each experiment was trained for 80 epochs at a learning rate of 0.003 on four Nvidia RTX3090 graphics cards. To demonstrate the performance of our dataset and the designed algorithms, we introduced some existing state-of-the-art algorithms, all carrying out experiments as required, we use five unimodal baseline models such as PointPillars⁴⁵, Voxel R-CNN⁴⁶, RDIoU⁴⁷, CasA-V⁴⁸, CasA-T⁴⁸ and two multi-modal baseline models such as VFF⁴⁹, M2-Fusion⁴⁰.

Data Records

The dataset is available in both the data repositories^50,51,52. Our image data is published as png images and Lidar point cloud and radar point cloud data is published as bin files, which are fully accessible to the user without any further licence. In this section, we describe the detailed contents and file directory of Dual Radar.

Dataset organization

We take three kinds of labels as the experiment objects, which are “Car”, “Pedestrian” and “Cyclist”. According to the need of the experiment, we split all the annotated data into a training, validation, and test set according to the ratio of 51%, 25%, and 24%. Our dataset will be published in the KITTI⁵³ format.

We conducted statistical analysis on our dataset and summarized the total counts for each label, as shown in Fig. 3. We presented a pie chart to display the object counts of the top six labels. The “Car” label slightly exceeds 50% of the total object count. Most of our data was obtained in urban conditions with decent traffic roads. As a result, the majority of labels are concentrated in “Car”, “Cyclist” and “Pedestrian”. We choose objects from these three labels to validate the performance of our dataset. We further analyzed the frame counts under different weather conditions and periods, as shown in Fig. 4 and Table 3. About two-thirds of our data were collected under normal weather conditions, and about one-third were collected under adverse weather. We collected 577 frames in adverse weather, which is about 5.5% of the total dataset. The adverse weather data we collect can be used to test the performance of different 4D radars in adverse weather conditions.

Table 3 Number of sequences of data in each scenario.

Full size table

We also collected data at dawn and night when low light intensity challenged the camera’s performance. We also conducted a statistical analysis of the number of objects with each label at different distance ranges from our vehicle, as shown in Fig. 5. Most objects are within 60 meters of our ego vehicle. The distribution of distances between “Bus”, “Truck” and our ego vehicle is uniform across each range. In addition, we analyzed the distribution density of the point clouds and the number of point clouds per frame, as shown in Fig. 6 and Table 4. In most cases, the LiDAR has 110K to 120K points per frame, the Arbe Phoenix radar has 6K to 14K points per frame, and the ARS548 RDI radar has 400 to 650 points per frame. Due to a lot of noise in the data collection set in the tunnel scenario, there will be part of the data with much larger than normal point clouds, as shown in 713 frames of LiDAR data with more than 130K point clouds and 158 frames of Arbe Phoenix radar data with more than 26K point clouds. Based on these statistical results, our dataset encompasses adverse driving scenarios, making it conducive for the application of objects like “Car”, “Pedestrian” and “Cyclist” in experiments. Moreover, the simultaneous collection of 4D radar point clouds from two types of 4D radar is useful for analyzing the effects of different 4D radar point clouds on specific driving scenarios and researching perceptual algorithms that can process different 4D radar point clouds. This dataset has significant implications for applying theoretical experiments in autonomous driving.

Table 4 The statistics of the number of point clouds per frame.

Full size table

Data Visualization

We visualize some of the data as shown in Figs. 7 and 8. Objects are annotated using 3D bounding boxes, which are mapped to the image, LiDAR point cloud, and two 4D radar point clouds. The 3D bounding box fits the object well and accurately depicts the corresponding points in the point cloud. Objects in both the LiDAR and 4D radar point clouds align well with those in the image, confirming good synchronization. The robustness of the 4D radar across different scenarios, weather conditions, and lighting conditions can be observed in Figs. 7 and 8. As shown in Fig. 8(g,h,i,j,k), the camera is highly affected by light and weather, and the camera image lacks RGB information to observe the object when the light is weak, backlit and adverse weather. In the corresponding scenarios, LiDAR acquires very tight spatial information, compensating for the camera’s shortcomings. However, the LiDAR cannot effectively distinguish overlapping or relatively close objects in scenarios with more objects. The Arbe Phoenix radar can collect a dense point cloud, which can collect all the objects within the field of view, but it contains much noise. The ARS548 RDI radar produces a less dense point cloud, with occasional detection leakage, but each point cloud accurately represents object information. The ARS548 RDI radar can also be used to collect the information of the object in the field of view.

Technical Validation

This section establishes the experimental platform for conducting the experiments on our dataset. We utilize several state-of-the-art baselines to validate our dataset. The performance of our dataset is verified by the results obtained from the experiments. We then conduct both qualitative and quantitative analyses on the results of the experiments, and finally, we get our evaluations of our dataset.

Experiments with our dataset in some single model

As shown in Table 5, the LiDAR point clouds captured in our dataset have excellent detection results, while the 4D radar point clouds have the potential for improvement. In the BEV view, the detection accuracies of the CasA-V model using the LiDAR point cloud reach 59.12% (Car), 49.35% (Pedestrian), and 50.03% (Cyclist), which indicates that the LiDAR point cloud in this dataset is able to represent the object information well. However, the detection accuracies are lower when using the 4D radar. In the BEV view, it can be observed that the CasA-V model achieves 12.28% and 3.74% in the “Car” category using the Arbe Phoenix and ARS548 RDI point clouds. This difference is due to the higher point density of the 80-line LiDAR, which collects much more information than the 4D radar. Meanwhile, the PointPillars model has achieved an accuracy of 35.09% in the category of “Car” when using the Arbe Phoenix radar point cloud, which is 26.95% higher than ARS548 RDI. This shows that despite the noise, the Arbe Phoenix radar point cloud can still represent the object information. The ARS548 RDI radar point cloud outperforms the Arbe Phoenix radar point cloud in the middle-sized “Cyclist” category. The accuracy of the PointPillars model on the ARS548 RDI radar point cloud reached 1.64% (Cyclist), which is 1.4% higher than the Arbe Phoenix radar point cloud. This is because cars with a large amount of metal are more conducive to collecting object information by LiDAR and radar. Although the Arbe Phoenix collects much noise, it can still better detect large metal objects. While the “Cyclist” object has less metal on it, collecting a lot of noise affects the model’s performance. The dataset label data defines the 3D bounding box through LiDAR point clouds, which cannot sometimes be captured by 4D radar because “pedestrian” objects usually have a smaller area, and 4D radar point clouds are sparser than LiDAR point clouds. The sparser nature of 4D radar point clouds compared to LiDAR point clouds can lead to situations where the radar does not detect the object, resulting in poor or even zero detection accuracy when training with 4D radar point clouds.

Table 5 Experimental results of single model baseline for three categories.

Full size table

Experiments with our dataset in some multi-modal

In addition, we discuss the Detection performance of the multi-modal baseline model using this dataset. As can be seen in Table 6, when the M²-Fusion model fuses LiDAR and 4D radar, the detection of the “Car” category has a huge performance improvement. The detection results in the BEV view are higher than those of the other single-modal baseline models and 30.24% higher than the model of PointPillars. This indicates that the 4D radar data of this dataset can provide auxiliary information for LiDAR. Meanwhile, in the 3D view of the “Cyclist” category, when the LiDAR and ARS548 RDI radar point clouds are fused, the result is 4.72% higher than the result of fusing LiDAR and Arbe Phoenix radar point clouds, and 0.6% higher than the CasA-V model using only LiDAR point clouds. This further demonstrates the superior detection performance of the ARS548 RDI radar point cloud for the “Cyclist” category. Moreover, the experimental results of the VFF model using the camera image fused with the LiDAR point cloud are Significantly outperform than most single-modal baseline models. In the BEV view, the VFF model using images and LiDAR has 25.16% higher detection results than the CasA-V model using only LiDAR in the “Car” category, which suggests that the camera can provide rich information and can perform well after fusing with the spatial information of the point cloud.

Table 6 Experimental results of mutiple modal baseline for three categories.

Full size table

Comparison of baseline models in adverse weather conditions

We compared the PointPillars and RDIoU models’ detection performance in adverse weather conditions. As can be seen in Table 7, the 4D radar performs better in adverse weather conditions. For the “Car” category, the RDIoU model has much higher detection results on radar point clouds than on LiDAR point clouds. The RDIoU model’s BEV and 3D view detection performance using the Arbe Phoenix radar point cloud is 14.78% and 10.27% better than LiDAR’s.

Table 7 Experimental results for three categories in the adverse weather scenario.

Full size table

However, for the “Cyclist” and “Pedestrian” categories, both 4D radar point clouds did not achieve satisfactory results. The PointPillars model achieved excellent results on the ARS548 RDI point cloud, while the RDIoU achieved excellent results on the Arbe Phoenix point cloud gave better results. These experimental results show that this dataset’s two 4D radar point clouds show different performance characteristics. This provides data support for further research on using 4D radar point clouds with different point cloud densities and noise levels in adverse scenarios and contributes to developing algorithms that can perform well on different 4D radar point clouds.

Categories: Electrical and electronic engineering, Mechanical engineering

Energy metabolism in health and diseases

Energy metabolism is indispensable for sustaining physiological functions in living organisms and assumes a pivotal role across physiological and pathological conditions. This review provides an extensive overview of advancements in energy metabolism research, elucidating critical pathways such as glycolysis, oxidative phosphorylation, fatty acid metabolism, and amino acid metabolism, along with their intricate regulatory mechanisms. The homeostatic balance of these processes is crucial; however, in pathological states such as neurodegenerative diseases, autoimmune disorders, and cancer, extensive metabolic reprogramming occurs, resulting in impaired glucose metabolism and mitochondrial dysfunction, which accelerate disease progression. Recent investigations into key regulatory pathways, including mechanistic target of rapamycin, sirtuins, and adenosine monophosphate-activated protein kinase, have considerably deepened our understanding of metabolic dysregulation and opened new avenues for therapeutic innovation. Emerging technologies, such as fluorescent probes, nano-biomaterials, and metabolomic analyses, promise substantial improvements in diagnostic precision. This review critically examines recent advancements and ongoing challenges in metabolism research, emphasizing its potential for precision diagnostics and personalized therapeutic interventions. Future studies should prioritize unraveling the regulatory mechanisms of energy metabolism and the dynamics of intercellular energy interactions. Integrating cutting-edge gene-editing technologies and multi-omics approaches, the development of multi-target pharmaceuticals in synergy with existing therapies such as immunotherapy and dietary interventions could enhance therapeutic efficacy. Personalized metabolic analysis is indispensable for crafting tailored treatment protocols, ultimately providing more accurate medical solutions for patients. This review aims to deepen the understanding and improve the application of energy metabolism to drive innovative diagnostic and therapeutic strategies.

AdMaPlace March 19, 2025

0 Comments

Tissue macrophages: origin, heterogenity, biological functions, diseases and therapeutic targets

Macrophages are immune cells belonging to the mononuclear phagocyte system. They play crucial roles in immune defense, surveillance, and homeostasis. This review systematically discusses the types of hematopoietic progenitors that give rise to macrophages, including primitive hematopoietic progenitors, erythro-myeloid progenitors, and hematopoietic stem cells. These progenitors have distinct genetic backgrounds and developmental processes. Accordingly, macrophages exhibit complex and diverse functions in the body, including phagocytosis and clearance of cellular debris, antigen presentation, and immune response, regulation of inflammation and cytokine production, tissue remodeling and repair, and multi-level regulatory signaling pathways/crosstalk involved in homeostasis and physiology. Besides, tumor-associated macrophages are a key component of the TME, exhibiting both anti-tumor and pro-tumor properties. Furthermore, the functional status of macrophages is closely linked to the development of various diseases, including cancer, autoimmune disorders, cardiovascular disease, neurodegenerative diseases, metabolic conditions, and trauma. Targeting macrophages has emerged as a promising therapeutic strategy in these contexts. Clinical trials of macrophage-based targeted drugs, macrophage-based immunotherapies, and nanoparticle-based therapy were comprehensively summarized. Potential challenges and future directions in targeting macrophages have also been discussed. Overall, our review highlights the significance of this versatile immune cell in human health and disease, which is expected to inform future research and clinical practice.

AdMaPlace March 19, 2025

0 Comments

Advance in peptide-based drug development: delivery platforms, therapeutics and vaccines

The successful approval of peptide-based drugs can be attributed to a collaborative effort across multiple disciplines. The integration of novel drug design and synthesis techniques, display library technology, delivery systems, bioengineering advancements, and artificial intelligence have significantly expedited the development of groundbreaking peptide-based drugs, effectively addressing the obstacles associated with their character, such as the rapid clearance and degradation, necessitating subcutaneous injection leading to increasing patient discomfort, and ultimately advancing translational research efforts. Peptides are presently employed in the management and diagnosis of a diverse array of medical conditions, such as diabetes mellitus, weight loss, oncology, and rare diseases, and are additionally garnering interest in facilitating targeted drug delivery platforms and the advancement of peptide-based vaccines. This paper provides an overview of the present market and clinical trial progress of peptide-based therapeutics, delivery platforms, and vaccines. It examines the key areas of research in peptide-based drug development through a literature analysis and emphasizes the structural modification principles of peptide-based drugs, as well as the recent advancements in screening, design, and delivery technologies. The accelerated advancement in the development of novel peptide-based therapeutics, including peptide-drug complexes, new peptide-based vaccines, and innovative peptide-based diagnostic reagents, has the potential to promote the era of precise customization of disease therapeutic schedule.

AdMaPlace March 19, 2025

0 Comments

Perceptual and semantic maps in individual humans share structural features that predict creative abilities

Building perceptual and associative links between internal representations is a fundamental neural process, allowing individuals to structure their knowledge about the world and combine it to enable efficient and creative behavior. In this context, the representational similarity between pairs of represented entities is thought to reflect their associative linkage at different levels of sensory processing, ranging from lower-order perceptual levels up to higher-order semantic levels. While recently specific structural features of semantic representational maps were linked with creative abilities of individual humans, it remains unclear if these features are also shared on lower level, perceptual maps. Here, we address this question by presenting 148 human participants with psychophysical scaling tasks, using two sets of independent and qualitatively distinct stimuli, to probe representational map structures in the lower-order auditory and the higher-order semantic domain. We quantify individual representational features with graph-theoretical measures and demonstrate a robust correlation of representational structures in the perceptual auditory and semantic modality. We delineate these shared representational features to predict multiple verbal standard measures of creativity, observing that both, semantic and auditory features, reflect creative abilities. Our findings indicate that the general, modality-overarching representational geometry of an individual is a relevant underpinning of creative thought.

AdMaPlace March 10, 2025

0 Comments

Probabilistic machine learning for battery health diagnostics and prognostics—review and perspectives

Diagnosing lithium-ion battery health and predicting future degradation is essential for driving design improvements in the laboratory and ensuring safe and reliable operation over a product’s expected lifetime. However, accurate battery health diagnostics and prognostics is challenging due to the unavoidable influence of cell-to-cell manufacturing variability and time-varying operating circumstances experienced in the field. Machine learning approaches informed by simulation, experiment, and field data show enormous promise to predict the evolution of battery health with use; however, until recently, the research community has focused on deterministic modeling methods, largely ignoring the cell-to-cell performance and aging variability inherent to all batteries. To truly make informed decisions regarding battery design in the lab or control strategies for the field, it is critical to characterize the uncertainty in a model’s predictions. After providing an overview of lithium-ion battery degradation, this paper reviews the current state-of-the-art probabilistic machine learning models for health diagnostics and prognostics. Details of the various methods, their advantages, and limitations are discussed in detail with a primary focus on probabilistic machine learning and uncertainty quantification. Last, future trends and opportunities for research and development are discussed.

AdMaPlace March 17, 2025

0 Comments

Background & Summary

Methods

Sensor Specification

Sensor Calibration

Dataset Annotation

Methods for technical validation

Data Records

Dataset organization

Data Visualization

Technical Validation

Experiments with our dataset in some single model

Experiments with our dataset in some multi-modal

Comparison of baseline models in adverse weather conditions

Related Articles

Responses