Richard — Capraru ~repack~

As he continues his Ph.D., Capraru’s future contributions are likely to be impactful. His unique blend of expertise in radar technology, machine learning, and cybersecurity positions him perfectly to develop holistic solutions for the next generation of autonomous systems. He is addressing the fundamental question of how machines can perceive and interact with the world safely, even when that world is trying to deceive them.

Before shifting fully into autonomous vehicle security, Dr. Capraru vastly expanded the open-source signal processing community's access to clean radar datasets. Alongside co-researchers from UCL and TU Delft, he developed .

While specific details about Richard Capraru’s early life are not widely publicized, his professional journey showcases a strong and focused academic background. The Capraru surname is most common in Romania, where it is believed to originate. However, his educational path began in the United Kingdom, where he attended .

: Utilizing deep learning and neural networks for signal, image, and video processing. richard capraru

Dr. Capraru’s academic path spans several elite global research hubs, combining foundations in hardware engineering with advanced artificial intelligence application:

: He explored the efficacy of affordable CW radar modules for gesture recognition

Autonomous perception frameworks rely heavily on LiDAR to map 3D environments using infrared light beams. However, atmospheric disruptions degrade these signals. In his foundational paper, “Upsampling Data Challenge: Object-Aware Approach for 3D Object Detection in Rain,” Dr. Capraru investigated how rain drops scatter light, attenuate intensity, and dramatically reduce the point cloud budget available to vehicle AI. He pioneered "object-aware" data approaches to reconstruct and "upsample" degraded sensory inputs, helping vehicles maintain high detection accuracy despite torrential downpours. 2. The Threat of Adversarial Sensor Attacks As he continues his Ph

Capraru has also explored a critical machine learning challenge known as where a model, when trained on a new task, loses the knowledge it previously learned. His work, "Overcoming Catastrophic Forgetting in Radar and Lidar Object Detection in Rain," proposes techniques like layer freezing and data augmentation to help autonomous systems maintain robust performance even in challenging, dynamically changing weather conditions.

: He has co-authored papers on using deep learning, specifically convolutional neural networks (CNNs), to count and localize people using 60 GHz FMCW radar. This includes addressing the resilience of these models in dynamic environments. Radar Data Challenges : Capraru was a contributor to the

Developing machine learning models for signal processing and image recognition. Key Scientific Contributions Before shifting fully into autonomous vehicle security, Dr

| Publication Title | Focus Area | Key Contribution | | :--- | :--- | :--- | | (2020) | Radar-based Gesture Recognition | Proved that low-cost Continuous Wave (CW) radar can match the gesture recognition accuracy of more complex systems. | | Dop-NET: a micro-Doppler radar data challenge (2020) | Radar Data & Machine Learning | Introduced a standard dataset to train machine learning algorithms for specific radar data. | | Exploring deep transfer learning interference classification... (2022) | Synthetic Data & SAR | Demonstrated that AI-generated synthetic radar data could be used to train other AI models effectively. | | Upsampling Data Challenge: Object-Aware Approach for 3D Object Detection in Rain (2023) | LiDAR & 3D Detection | Proposed a new data processing method to improve object detection for autonomous vehicles in rainy conditions. | | Rain-Reaper: Unmasking LiDAR-based Detector Vulnerabilities in Rain (2024, IROS) | LiDAR Security & Weather | Developed an attack that exploits rain’s physical properties to trick a LiDAR system into ignoring real obstacles. | | Leveraging Adverse Weather for Enhanced LiDAR Spoofing... (2026, IEEE Vehicular Technology Magazine ) | Autonomous Vehicle Security | Argued that weather isn't just a hindrance but can be strategically leveraged to design more sophisticated attacks on self-driving car sensors. |

The Capraru Continuum argues for the "Sweet Spot" in the middle: . This approach retains the spatial logic and structural markers of the industrial past (crane tracks, silos, high-bay ceilings) while inserting distinct, autonomous modern volumes within them. This creates a visual friction that heightens the experience of both the old and the new.

: He has investigated the vulnerabilities of 3D object detection systems, specifically looking at how physical adversaries can spoof LiDAR signals to create "ghost objects". Radar-Based Gesture Recognition

Expanded his academic scope as a visiting scholar across East Asia's top institutions, including Peking University, the Hong Kong University of Science and Technology, Korea University, and the University of Tokyo. Key Research Breakthroughs 1. Unmasking LiDAR Vulnerabilities in Bad Weather