Georg Stettinger
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Georg Stettinger has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
- Manuel Pegalajar Cuellar
Department of Computer Science and Artificial Intelligence, University of Granada, 18014 Granada, Spain
5 shared publications - Miguel Molina-Solana
Department of Computer Science and Artificial Intelligence, University of Granada, 18071 Granada, Spain
3 shared publications - Vasu Tammisetti
Infineon Technologies AG, 85579 Munich, Germany
3 shared publications - Diego Pedro Morales
Department of Electronics and Computer Technology, University of Granada, 18071 Granada, Spain
2 shared publications - Elfi Fertl
Infineon Technologies AG, 85579 Neubiberg, Germany
2 shared publications - Encarnación Castillo
Universidad de Granada
2 shared publications
Latest Publications
- Open Access
- 11 November 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (22) , 11964
Real-Time Detection of Rear Car Signals for Advanced Driver Assistance Systems Using Meta-Learning and Geometric Post-Processing
Accurate identification of rear light signals in preceding vehicles is pivotal for Advanced Driver Assistance Systems (ADAS), enabling early detection of driver intentions and thereby improving road safety. In this work, we present a novel approach that leverages a meta-learning-enhanced YOLOv8 model to detect left and right turn indicators, as well as brake signals. Traditional radar and LiDAR provide robust geometry, range, and motion cues that can indirectly suggest driver intent (e.g., deceleration or lane drift). However, they do not directly interpret color-coded rear signals, which limits early intent recognition from the taillights. We therefore focus on a camera-based approach that complements ranging sensors by decoding color and spatial patterns in rear lights. This approach to detecting vehicle signals poses additional challenges due to factors such as high reflectivity and the subtle visual differences between directional indicators. We address these by training a YOLOv8 model with a meta-learning strategy, thus enhancing its capability to learn from minimal data and rapidly adapt to new scenarios. Furthermore, we developed a post-processing layer that classifies signals by the geometric properties of detected objects, employing mathematical principles such as distance, area calculation, and Intersection over Union (IoU) metrics. Our approach increases adaptability and performance compared to traditional deep learning techniques, supporting the conclusion that integrating meta-learning into real-time object detection frameworks provides a scalable and robust solution for intelligent vehicle perception, significantly enhancing situational awareness and road safety through reliable prediction of vehicular behavior.
- 11 November 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (22) , 11964
Open AccessReal-Time Detection of Rear Car Signals for Advanced Driver Assistance Systems Using Meta-Learning and Geometric Post-Processing
Accurate identification of rear light signals in preceding vehicles is pivotal for Advanced Driver Assistance Systems (ADAS), enabling early detection of driver intentions and thereby improving road safety. In this work, we present a novel approach that leverages a meta-learning-enhanced YOLOv8 model to detect left and right turn indicators, as well as brake signals. Traditional radar and LiDAR provide robust geometry, range, and motion cues that can indirectly suggest driver intent (e.g., deceleration or lane drift). However, they do not directly interpret color-coded rear signals, which limits early intent recognition from the taillights. We therefore focus on a camera-based approach that complements ranging sensors by decoding color and spatial patterns in rear lights. This approach to detecting vehicle signals poses additional challenges due to factors such as high reflectivity and the subtle visual differences between directional indicators. We address these by training a YOLOv8 model with a meta-learning strategy, thus enhancing its capability to learn from minimal data and rapidly adapt to new scenarios. Furthermore, we developed a post-processing layer that classifies signals by the geometric properties of detected objects, employing mathematical principles such as distance, area calculation, and Intersection over Union (IoU) metrics. Our approach increases adaptability and performance compared to traditional deep learning techniques, supporting the conclusion that integrating meta-learning into real-time object detection frameworks provides a scalable and robust solution for intelligent vehicle perception, significantly enhancing situational awareness and road safety through reliable prediction of vehicular behavior.
- Open Access
What's Really Different with AI? -- A Behavior-based Perspective on System Safety for Automated Driving Systems
Assuring safety for ``AI-based'' systems is one of the current challenges in safety engineering. For automated driving systems, in particular, further assurance challenges result from the open context that the systems need to operate in after deployment. The current standardization and regulation landscape for ``AI-based'' systems is becoming ever more complex, as standards and regulations are being released at high frequencies. This position paper seeks to provide guidance for making qualified arguments which standards should meaningfully be applied to (``AI-based'') automated driving systems. Furthermore, we argue for clearly differentiating sources of risk between AI-specific and general uncertainties related to the open context. In our view, a clear conceptual separation can help to exploit commonalities that can close the gap between system-level and AI-specific safety analyses, while ensuring the required rigor for engineering safe ``AI-based'' systems.
Open AccessWhat's Really Different with AI? -- A Behavior-based Perspective on System Safety for Automated Driving Systems
Assuring safety for ``AI-based'' systems is one of the current challenges in safety engineering. For automated driving systems, in particular, further assurance challenges result from the open context that the systems need to operate in after deployment. The current standardization and regulation landscape for ``AI-based'' systems is becoming ever more complex, as standards and regulations are being released at high frequencies. This position paper seeks to provide guidance for making qualified arguments which standards should meaningfully be applied to (``AI-based'') automated driving systems. Furthermore, we argue for clearly differentiating sources of risk between AI-specific and general uncertainties related to the open context. In our view, a clear conceptual separation can help to exploit commonalities that can close the gap between system-level and AI-specific safety analyses, while ensuring the required rigor for engineering safe ``AI-based'' systems.
- Open Access
- 16 June 2025
- Published by Elsevier in IFAC Journal of Systems and Control
- Vol. 33
Exploring the potential of standardized behaviour competencies in automated driving systems
- 16 June 2025
- Published by Elsevier in IFAC Journal of Systems and Control
- Vol. 33
Open AccessExploring the potential of standardized behaviour competencies in automated driving systems
- Open Access
- 21 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (7) , 7007-7014
CaRaFFusion: Improving 2D Semantic Segmentation With Camera-Radar Point Cloud Fusion and Zero-Shot Image Inpainting
Segmenting objects in an environment is a crucial task for autonomous driving and robotics, as it enables a better understanding of the surroundings of each agent. Although camera sensors provide rich visual details, they are vulnerable to adverse weather conditions. In contrast, radar sensors remain robust under such conditions, but often produce sparse and noisy data. Therefore, a promising approach is to fuse information from both sensors. In this work, we propose a novel framework to enhance camera-only baselines by integrating a diffusion model into a camera-radar fusion architecture. We leverage radar point features to create pseudo-masks using the Segment-Anything model, treating the projected radar points as point prompts. Additionally, we propose a noise reduction unit to denoise these pseudo-masks, which are further used to generate inpainted images that complete the missing information in the original images. Our method improves the camera-only segmentation baseline by 2.63% in mIoU and enhances our camera-radar fusion architecture by 1.48% in mIoU on the Waterscenes dataset. This demonstrates the effectiveness of our approach for semantic segmentation using camera-radar fusion under adverse weather conditions.
- 21 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (7) , 7007-7014
Open AccessCaRaFFusion: Improving 2D Semantic Segmentation With Camera-Radar Point Cloud Fusion and Zero-Shot Image Inpainting
Segmenting objects in an environment is a crucial task for autonomous driving and robotics, as it enables a better understanding of the surroundings of each agent. Although camera sensors provide rich visual details, they are vulnerable to adverse weather conditions. In contrast, radar sensors remain robust under such conditions, but often produce sparse and noisy data. Therefore, a promising approach is to fuse information from both sensors. In this work, we propose a novel framework to enhance camera-only baselines by integrating a diffusion model into a camera-radar fusion architecture. We leverage radar point features to create pseudo-masks using the Segment-Anything model, treating the projected radar points as point prompts. Additionally, we propose a noise reduction unit to denoise these pseudo-masks, which are further used to generate inpainted images that complete the missing information in the original images. Our method improves the camera-only segmentation baseline by 2.63% in mIoU and enhances our camera-radar fusion architecture by 1.48% in mIoU on the Waterscenes dataset. This demonstrates the effectiveness of our approach for semantic segmentation using camera-radar fusion under adverse weather conditions.
- Top Cited PapersOpen Access
Hand Gesture Recognition on Edge Devices: Sensor Technologies, Algorithms, and Processing Hardware
Hand gesture recognition (HGR) is a convenient and natural form of human–computer interaction. It is suitable for various applications. Much research has already focused on wearable device-based HGR. By contrast, this paper gives an overview focused on device-free HGR. That means we evaluate HGR systems that do not require the user to wear something like a data glove or hold a device. HGR systems are explored regarding technology, hardware, and algorithms. The interconnectedness of timing and power requirements with hardware, pre-processing algorithm, classification, and technology and how they permit more or less granularity, accuracy, and number of gestures is clearly demonstrated. Sensor modalities evaluated are WIFI, vision, radar, mobile networks, and ultrasound. The pre-processing technologies stereo vision, multiple-input multiple-output (MIMO), spectrogram, phased array, range-doppler-map, range-angle-map, doppler-angle-map, and multilateration are explored. Classification approaches with and without ML are studied. Among those with ML, assessed algorithms range from simple tree structures to transformers. All applications are evaluated taking into account their level of integration. This encompasses determining whether the application presented is suitable for edge integration, their real-time capability, whether continuous learning is implemented, which robustness was achieved, whether ML is applied, and the accuracy level. Our survey aims to provide a thorough understanding of the current state of the art in device-free HGR on edge devices and in general. Finally, on the basis of present-day challenges and opportunities in this field, we outline which further research we suggest for HGR improvement. Our goal is to promote the development of efficient and accurate gesture recognition systems.
Top Cited PapersOpen AccessHand Gesture Recognition on Edge Devices: Sensor Technologies, Algorithms, and Processing Hardware
Hand gesture recognition (HGR) is a convenient and natural form of human–computer interaction. It is suitable for various applications. Much research has already focused on wearable device-based HGR. By contrast, this paper gives an overview focused on device-free HGR. That means we evaluate HGR systems that do not require the user to wear something like a data glove or hold a device. HGR systems are explored regarding technology, hardware, and algorithms. The interconnectedness of timing and power requirements with hardware, pre-processing algorithm, classification, and technology and how they permit more or less granularity, accuracy, and number of gestures is clearly demonstrated. Sensor modalities evaluated are WIFI, vision, radar, mobile networks, and ultrasound. The pre-processing technologies stereo vision, multiple-input multiple-output (MIMO), spectrogram, phased array, range-doppler-map, range-angle-map, doppler-angle-map, and multilateration are explored. Classification approaches with and without ML are studied. Among those with ML, assessed algorithms range from simple tree structures to transformers. All applications are evaluated taking into account their level of integration. This encompasses determining whether the application presented is suitable for edge integration, their real-time capability, whether continuous learning is implemented, which robustness was achieved, whether ML is applied, and the accuracy level. Our survey aims to provide a thorough understanding of the current state of the art in device-free HGR on edge devices and in general. Finally, on the basis of present-day challenges and opportunities in this field, we outline which further research we suggest for HGR improvement. Our goal is to promote the development of efficient and accurate gesture recognition systems.
- Open Access
- 24 January 2025
- Published by MDPI AG in Electronics
- Vol. 14 (3) , 468
Meta-YOLOv8: Meta-Learning-Enhanced YOLOv8 for Precise Traffic Light Color Detection in ADAS
The ability to accurately detect traffic light color is critical for the functioning of Advanced Driver Assistance Systems (ADAS), as it directly impacts a vehicle’s safety and operational efficiency. This paper introduces Meta-YOLOv8, an improvement over YOLOv8 based on meta-learning, designed explicitly for traffic light color detection focusing on color recognition. In contrast to conventional models, Meta-YOLOv8 focuses on the illuminated portion of traffic signals, enhancing accuracy and extending the detection range in challenging conditions. Furthermore, this approach reduces the computational load by filtering out irrelevant data. An innovative labeling technique has been implemented to address real-time weather-related detection issues, although other bright objects may occasionally confound it. Our model employs meta-learning principles to mitigate confusion and boost confidence in detections. Leveraging task similarity and prior knowledge enhances detection performance across diverse lighting and weather conditions. Meta-learning also reduces the necessity for extensive datasets while maintaining consistent performance and adaptability to novel categories. The optimized feature weighting for precise color differentiation, coupled with reduced latency and computational demands, enables a faster response from the driver and reduces the risk of accidents. This represents a significant advancement for resource-constrained ADAS. A comparative assessment of Meta-YOLOv8 with traditional models, including SSD, Faster R-CNN, and Detection Transformers (DETR), reveals that it outperforms these models, achieving an F1 score, accuracy of 93% and a precision rate of 97%.
- 24 January 2025
- Published by MDPI AG in Electronics
- Vol. 14 (3) , 468
Open AccessMeta-YOLOv8: Meta-Learning-Enhanced YOLOv8 for Precise Traffic Light Color Detection in ADAS
The ability to accurately detect traffic light color is critical for the functioning of Advanced Driver Assistance Systems (ADAS), as it directly impacts a vehicle’s safety and operational efficiency. This paper introduces Meta-YOLOv8, an improvement over YOLOv8 based on meta-learning, designed explicitly for traffic light color detection focusing on color recognition. In contrast to conventional models, Meta-YOLOv8 focuses on the illuminated portion of traffic signals, enhancing accuracy and extending the detection range in challenging conditions. Furthermore, this approach reduces the computational load by filtering out irrelevant data. An innovative labeling technique has been implemented to address real-time weather-related detection issues, although other bright objects may occasionally confound it. Our model employs meta-learning principles to mitigate confusion and boost confidence in detections. Leveraging task similarity and prior knowledge enhances detection performance across diverse lighting and weather conditions. Meta-learning also reduces the necessity for extensive datasets while maintaining consistent performance and adaptability to novel categories. The optimized feature weighting for precise color differentiation, coupled with reduced latency and computational demands, enables a faster response from the driver and reduces the risk of accidents. This represents a significant advancement for resource-constrained ADAS. A comparative assessment of Meta-YOLOv8 with traditional models, including SSD, Faster R-CNN, and Detection Transformers (DETR), reveals that it outperforms these models, achieving an F1 score, accuracy of 93% and a precision rate of 97%.
- Open Access
- 22 November 2024
- Published by MDPI AG in Electronics
- Vol. 13 (23) , 4606
Explainable Safety Argumentation for the Deployment of Automated Vehicles
With over 1.6 million traffic deaths in 2016, automated vehicles equipped with automated driving systems (ADSs) have the potential to increase traffic safety by assuming human driving tasks within the operational design domain (ODD). However, safety validation is challenging due to the open-context problem. Current strategies, such as pure driving and requirement-based testing, are insufficient. Scenario-based testing offers a solution but necessitates appropriate scenario selection, testing methods, and evaluation criteria. This paper builds upon a method to calculate the covered ODD using tested scenarios generated from logical scenarios, considering parameter discretisation uncertainty. Acceptance criteria for the safety argumentation are proposed based on parameter space coverage and variance introduced via discretisation, thus contributing to quantifying the residual risks of safety validation. The approach is demonstrated through two logical scenarios with probability density functions of the parameters generated using a trajectory dataset. These criteria can serve as risk acceptance criteria, providing comparability and explainable results. By developing a robust scenario-based testing approach, ADS safety can be validated, leading to increased traffic safety and reduced fatalities. Since ADSs incorporate AI models, this proposed validation strategy can be extended to AI systems across multiple domains for the respective assurance argument required for deployment.
- 22 November 2024
- Published by MDPI AG in Electronics
- Vol. 13 (23) , 4606
Open AccessExplainable Safety Argumentation for the Deployment of Automated Vehicles
With over 1.6 million traffic deaths in 2016, automated vehicles equipped with automated driving systems (ADSs) have the potential to increase traffic safety by assuming human driving tasks within the operational design domain (ODD). However, safety validation is challenging due to the open-context problem. Current strategies, such as pure driving and requirement-based testing, are insufficient. Scenario-based testing offers a solution but necessitates appropriate scenario selection, testing methods, and evaluation criteria. This paper builds upon a method to calculate the covered ODD using tested scenarios generated from logical scenarios, considering parameter discretisation uncertainty. Acceptance criteria for the safety argumentation are proposed based on parameter space coverage and variance introduced via discretisation, thus contributing to quantifying the residual risks of safety validation. The approach is demonstrated through two logical scenarios with probability density functions of the parameters generated using a trajectory dataset. These criteria can serve as risk acceptance criteria, providing comparability and explainable results. By developing a robust scenario-based testing approach, ADS safety can be validated, leading to increased traffic safety and reduced fatalities. Since ADSs incorporate AI models, this proposed validation strategy can be extended to AI systems across multiple domains for the respective assurance argument required for deployment.
- Open Access
End-to-End Ultrasonic Hand Gesture Recognition
As the number of electronic gadgets in our daily lives is increasing and most of them require some kind of human interaction, this demands innovative, convenient input methods. There are limitations to state-of-the-art (SotA) ultrasound-based hand gesture recognition (HGR) systems in terms of robustness and accuracy. This research presents a novel machine learning (ML)-based end-to-end solution for hand gesture recognition with low-cost micro-electromechanical (MEMS) system ultrasonic transducers. In contrast to prior methods, our ML model processes the raw echo samples directly instead of using pre-processed data. Consequently, the processing flow presented in this work leaves it to the ML model to extract the important information from the echo data. The success of this approach is demonstrated as follows. Four MEMS ultrasonic transducers are placed in three different geometrical arrangements. For each arrangement, different types of ML models are optimized and benchmarked on datasets acquired with the presented custom hardware (HW): convolutional neural networks (CNNs), gated recurrent units (GRUs), long short-term memory (LSTM), vision transformer (ViT), and cross-attention multi-scale vision transformer (CrossViT). The three last-mentioned ML models reached more than 88% accuracy. The most important innovation described in this research paper is that we were able to demonstrate that little pre-processing is necessary to obtain high accuracy in ultrasonic HGR for several arrangements of cost-effective and low-power MEMS ultrasonic transducer arrays. Even the computationally intensive Fourier transform can be omitted. The presented approach is further compared to HGR systems using other sensor types such as vision, WiFi, radar, and state-of-the-art ultrasound-based HGR systems. Direct processing of the sensor signals by a compact model makes ultrasonic hand gesture recognition a true low-cost and power-efficient input method.
Open AccessEnd-to-End Ultrasonic Hand Gesture Recognition
As the number of electronic gadgets in our daily lives is increasing and most of them require some kind of human interaction, this demands innovative, convenient input methods. There are limitations to state-of-the-art (SotA) ultrasound-based hand gesture recognition (HGR) systems in terms of robustness and accuracy. This research presents a novel machine learning (ML)-based end-to-end solution for hand gesture recognition with low-cost micro-electromechanical (MEMS) system ultrasonic transducers. In contrast to prior methods, our ML model processes the raw echo samples directly instead of using pre-processed data. Consequently, the processing flow presented in this work leaves it to the ML model to extract the important information from the echo data. The success of this approach is demonstrated as follows. Four MEMS ultrasonic transducers are placed in three different geometrical arrangements. For each arrangement, different types of ML models are optimized and benchmarked on datasets acquired with the presented custom hardware (HW): convolutional neural networks (CNNs), gated recurrent units (GRUs), long short-term memory (LSTM), vision transformer (ViT), and cross-attention multi-scale vision transformer (CrossViT). The three last-mentioned ML models reached more than 88% accuracy. The most important innovation described in this research paper is that we were able to demonstrate that little pre-processing is necessary to obtain high accuracy in ultrasonic HGR for several arrangements of cost-effective and low-power MEMS ultrasonic transducer arrays. Even the computationally intensive Fourier transform can be omitted. The presented approach is further compared to HGR systems using other sensor types such as vision, WiFi, radar, and state-of-the-art ultrasound-based HGR systems. Direct processing of the sensor signals by a compact model makes ultrasonic hand gesture recognition a true low-cost and power-efficient input method.
- Open Access
- 22 April 2024
- Published by MDPI AG in Electronics
- Vol. 13 (8) , 1585
LaANIL: ANIL with Look-Ahead Meta-Optimization and Data Parallelism
Meta-few-shot learning algorithms, such as Model-Agnostic Meta-Learning (MAML) and Almost No Inner Loop (ANIL), enable machines to learn complex tasks quickly with limited data and based on previous experience. By maintaining the inner loop head of the neural network, ANIL leads to simpler computations and reduces the complexity of MAML. Despite its benefits, ANIL suffers from issues like accuracy variance, slow initial learning, and overfitting, hardening its adaptation and generalization. This work proposes “Look-Ahead ANIL” (LaANIL), an enhancement to ANIL for better learning. LaANIL reorganizes ANIL’s internal architecture, integrating parallel computing techniques (to process multiple training examples simultaneously across computing units) and incorporating Nesterov momentum (which accelerates convergence by adjusting the learning rate based on past gradient information and extracting informative features for look-ahead gradient computation). These additional features make our model more state-of-the-art capable and better edge-compatible and thus improve few-short learning by enabling models to quickly adapt to new information and tasks. LaANIL’s effectiveness is validated on established meta-few-shot learning datasets, including FC100, CIFAR-FS, Mini-ImageNet, CUBirds-200-2011, and Tiered-ImageNet. The proposed model achieved an increased validation accuracy by 7 ± 0.7% and a variance reduction by 44 ± 4% in two-way two-shot classification as well as increased validation by 5 ± 0.4% and a variance reduction by 18 ± 2% in five-way five-shot classification on the FC100 dataset and similarly performed well on other datasets.
- 22 April 2024
- Published by MDPI AG in Electronics
- Vol. 13 (8) , 1585
Open AccessLaANIL: ANIL with Look-Ahead Meta-Optimization and Data Parallelism
Meta-few-shot learning algorithms, such as Model-Agnostic Meta-Learning (MAML) and Almost No Inner Loop (ANIL), enable machines to learn complex tasks quickly with limited data and based on previous experience. By maintaining the inner loop head of the neural network, ANIL leads to simpler computations and reduces the complexity of MAML. Despite its benefits, ANIL suffers from issues like accuracy variance, slow initial learning, and overfitting, hardening its adaptation and generalization. This work proposes “Look-Ahead ANIL” (LaANIL), an enhancement to ANIL for better learning. LaANIL reorganizes ANIL’s internal architecture, integrating parallel computing techniques (to process multiple training examples simultaneously across computing units) and incorporating Nesterov momentum (which accelerates convergence by adjusting the learning rate based on past gradient information and extracting informative features for look-ahead gradient computation). These additional features make our model more state-of-the-art capable and better edge-compatible and thus improve few-short learning by enabling models to quickly adapt to new information and tasks. LaANIL’s effectiveness is validated on established meta-few-shot learning datasets, including FC100, CIFAR-FS, Mini-ImageNet, CUBirds-200-2011, and Tiered-ImageNet. The proposed model achieved an increased validation accuracy by 7 ± 0.7% and a variance reduction by 44 ± 4% in two-way two-shot classification as well as increased validation by 5 ± 0.4% and a variance reduction by 18 ± 2% in five-way five-shot classification on the FC100 dataset and similarly performed well on other datasets.
- Open Access
- 8 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 22718-22745
Trustworthiness Assurance Assessment for High-Risk AI-Based Systems
This work proposes methodologies for ensuring the trustworthiness of high-risk artificial intelligence (AI) systems (AIS) to achieve compliance with the European Union’s (EU) AI Act. High-risk classified AIS must fulfill seven requirements to be considered trustworthy and human-centric, and subsequently be considered for deployment. These requirements are equally important, mutually supportive, and should be implemented and evaluated throughout the AI lifecycle. The assurance of trustworthiness is influenced by ethical considerations, amongst others. Hence, the operational design domain (ODD) and behavior competency (BC) concepts from the automated driving domain are utilized in risk assessment strategies to quantify different types of residual risks. The methodology presented is guided by the consistent application of the ODD and its related BC concept throughout the entire AI lifecycle, focusing on the trustworthiness assurance framework and its associated process as the main pillars for AIS certification. The achievement of the overall objective of trustworthy and human-centric AIS is divided into seven interconnected sub-goals: the formulation of use restrictions, the trustworthiness assurance/argument itself, the identification of dysfunctional cases, the utilization of scenario databases and datasets, the application of metrics for evaluation, the implementation of the proposed concept across the AI lifecycle, and sufficient consideration of human factors. The role of standards in the assurance process is discussed, considering any existing gaps and areas for improvement. The work concludes with a summary of the developed approach, highlighting key takeaways and action points. Finally, a roadmap to ensure trustworthy and human-centric behavior of future AIS is outlined.
- 8 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 22718-22745
Open AccessTrustworthiness Assurance Assessment for High-Risk AI-Based Systems
This work proposes methodologies for ensuring the trustworthiness of high-risk artificial intelligence (AI) systems (AIS) to achieve compliance with the European Union’s (EU) AI Act. High-risk classified AIS must fulfill seven requirements to be considered trustworthy and human-centric, and subsequently be considered for deployment. These requirements are equally important, mutually supportive, and should be implemented and evaluated throughout the AI lifecycle. The assurance of trustworthiness is influenced by ethical considerations, amongst others. Hence, the operational design domain (ODD) and behavior competency (BC) concepts from the automated driving domain are utilized in risk assessment strategies to quantify different types of residual risks. The methodology presented is guided by the consistent application of the ODD and its related BC concept throughout the entire AI lifecycle, focusing on the trustworthiness assurance framework and its associated process as the main pillars for AIS certification. The achievement of the overall objective of trustworthy and human-centric AIS is divided into seven interconnected sub-goals: the formulation of use restrictions, the trustworthiness assurance/argument itself, the identification of dysfunctional cases, the utilization of scenario databases and datasets, the application of metrics for evaluation, the implementation of the proposed concept across the AI lifecycle, and sufficient consideration of human factors. The role of standards in the assurance process is discussed, considering any existing gaps and areas for improvement. The work concludes with a summary of the developed approach, highlighting key takeaways and action points. Finally, a roadmap to ensure trustworthy and human-centric behavior of future AIS is outlined.