News

The Horizon Europe project BERTHA (BEhavioural Replication of Human drivers for CCAM) presented its results at its final event on 21 October 2026 in Brussels. The event brought together researchers, industry, policymakers and mobility stakeholders to review the project’s achievements and discuss the next steps towards human-centred automated mobility in Europe.

Dr. Jason Rambach presented the perception module developed by the Augmented Vision department. The module models where drivers look and which objects they attend to while driving, and forms part of BERTHA’s Driver Behavioural Model.

The project outcomes from DFKI AV include: DriverGaze360, a large-scale driver attention dataset with a full 360-degree field of view, containing around 1 million gaze-labelled frames recorded from 19 drivers, and DriverGaze360-Net, a panoramic attention prediction network that jointly predicts attention maps and attended objects. The approach achieves state-of-the-art attention prediction on panoramic driving images and was published at CVPR 2026. Code and dataset are openly available.

Project page, code and dataset: https://dfki-av.github.io/drivergaze360/

BERTHA brought together 14 partners from 6 countries with more than €7.9 million in EU funding to develop a probabilistic Driver Behavioural Model that makes connected automated vehicles safer, more predictable and more human-like. The model combines modules for perception, risk awareness, decision-making, affect and motor behaviour, and is made available through an open-source data hub and the CARLA simulator. We thank all partners for three years of excellent collaboration.

Contact: Dr. Jason Rambach

The GEIHSR project has officially started! During the kick-off meeting held at DFKI in Kaiserslautern, all partners came together to align on the project’s objectives, technical approach, and first steps.

GEIHSR aims to advance safe, affordable, and human-safe industrial robotics by integrating AI-based perception, Robot Foundation Models, and efficient edge AI. The project seeks to enable robots to understand dynamic industrial environments, anticipate human intentions, and act safely and adaptively.

The project is a German-Taiwanese collaboration that will develop and validate an innovative robotics system, helping bring advanced automation closer to the needs of small and medium-sized enterprises.

Duration: 01.09.2026 – 31.08.2029
Funded by: Federal Ministry of Research, Technology, and Space
Coordination: DFKI
Partners: National Taiwan University, EngRoTec GmbH & Co. KG, Solectrix GmbH

On 14 October 2026, Mohammad Mahdi Chamseddine successfully defended his doctoral thesis entitled “Multi-Sensor Point Cloud Learning for 3D Scene Perception and Modelling”.

The doctoral thesis was written in the “Augmented Vision” department at DFKI under the supervision of Prof. Dr Didier Stricker and Dr.-Ing. Jason Raphael Rambach.

The examination board/examiners consisted of Prof. Dr Didier Stricker (RPTU), Prof. Dr Sophie Fellenz (RPTU), Prof. Dr Moses Ganardi (RPTU) and Prof. Dr Benjamin Busam (TUM).

Shashank Mishra and René Schuster from the Augmented Vision department at DFKI have been recognized as a Top Reviewers at NeurIPS 2026, one of the leading international conferences in machine learning and artificial intelligence.

This recognition acknowledges their contribution to the scientific peer-review process through high-quality and constructive reviews, supporting the evaluation and advancement of research in machine learning.

Contact: Shashank Mishra

Our paper, “Conditional Attribution for Root Cause Analysis in Time-Series Anomaly Detection,” by Shashank Mishra, Karan Patil, Cedric Schockaert, Didier Stricker, and Jason Rambach, was accepted for an oral presentation at ECML PKDD 2026, held in Naples, Italy, from September 7 to 11, 2026. Shashank Mishra presented the work at the conference.

The paper introduces a novel conditional attribution framework for identifying the sensors responsible for anomalies in multivariate time-series data. By comparing anomalous observations with contextually similar normal system states, the proposed approach preserves dependencies between sensors and provides more reliable explanations. Experiments on industrial benchmark datasets demonstrate improvements in root cause identification, temporal localization, and robustness.

Paper: https://link.springer.com/chapter/10.1007/978-3-032-37685-5_10

Code: https://github.com/dfki-av/Conditional-Attribution-for-Root-Cause-Analysis-in-Time-Series-Anomaly-Detection

Contact: Shashank Mishra

The dAIEDGE consortium held its 6th and final General Assembly in Modena, Italy. The two-day meeting was hosted by the University of Modena and Reggio Emilia and brought together the project partners to review the latest achievements, discuss the remaining activities and prepare for the final phase of the project.

dAIEDGE is coordinated by Dr. Alain Pagani from the Augmented Vision department at DFKI. During the meeting, the consortium reviewed progress across the project work packages, including technical developments, dissemination and exploitation activities, strategic research priorities, and preparation for the final review. DFKI led the sessions on WP2 and, together with the University of Edinburgh, WP1.

A further focus of the meeting was the preparation of the dAIEDGE Innovation Days, which will take place on 21–22 October 2026. The event will present the technologies, results and impact developed throughout the project and bring the dAIEDGE community together for the final project event.

The General Assembly also included a visit to MASA’s facilities, where participants had the opportunity to see dAIEDGE use cases in action and connect the project’s research and technological developments with real-world applications.

The meeting in Modena marked an important milestone for the consortium, providing an opportunity to review the progress achieved over the course of the project and to align the final activities before the conclusion of dAIEDGE.

Contact: Dr. Alain Pagani

Dr. Jason Rambach and Dr. Alain Pagani from the Augmented Vision department have been selected as Outstanding Reviewers at the European Conference on Computer Vision, ECCV 2026.

The distinction recognizes reviewers who provided particularly high-quality and valuable reviews during the conference reviewing process. Being selected as an Outstanding Reviewer reflects the researchers’ contribution to maintaining the scientific quality of one of the leading international conferences in computer vision.

The Augmented Vision department congratulates Dr. Jason Rambach and Dr. Alain Pagani on this recognition.

Contact: Dr. Alain Pagani

              Dr. Jason Rambach

The Augmented Vision department participated in the “Fifth Hands-on Egocentric Research Tutorial” at ECCV 2026 in Malmö, Sweden, on September 8, 2026. The tutorial was organised by Meta Project Aria and focused on research using egocentric sensing and wearable devices.

Dr. Alain Pagani gave the talk “From Hands to 3D: Egocentric Perception with Aria”, presenting recent research of the Augmented Vision department on 3D hand perception, hand-object interaction and 3D scene reconstruction from egocentric cameras.

The presentation included EgoForce, the department’s recent work on metric 3D hand pose and shape estimation from a monocular egocentric camera, as well as ongoing research on contact and force understanding in hand-object interactions. It also covered 3D scene reconstruction from Aria glasses using Gaussian Splatting, including reconstruction from very sparse image observations.

The tutorial provided an opportunity to present the department’s current work in egocentric perception and to exchange with the Project Aria research community.

Contact: Dr. Alain Pagani

We are pleased to announce that our paper, “Sequence-SOD: Bio-inspired Sequence-aware Spiking Object Detection for Event Cameras,” by Katharina Bendig, René Schuster, and Didier Stricker, has been accepted for publication in Cognitive Computation, a Springer Nature journal. The paper introduces a sequence-aware approach that enables spiking neural networks to better exploit temporal information in continuous event-camera data.

Read the paper

Contact: René Schuster

We are pleased to announce that DFKI’s Augmented Vision department was invited to Meta Reality Labs in Redmond, WA, to present our Project Aria work at the third Egocentric Intelligence Research Summit. Project Aria is Meta’s research platform for egocentric perception; display-free sensor glasses that let academic partners collect first-person data for work in machine perception, contextual AI, and augmented reality.

Shreedhar Govil and Shaoxiang Wang represented the department, giving a talk and poster presentation on DriverGaze360 and EgoForce. We look forward to continuing our collaboration with Meta and thank them for access to this data collection tool and research community.

Contact: shreedhar.govil@dfki.de and shaoxiang.wang@dfki.de  

The ENFIELD Final Conference, “Advancing AI in Europe: From Foundations to Real-World Applications,” brought together researchers, policymakers, industry representatives and European AI initiatives in Brussels on 7 and 8 July 2026. As a sister Network of Excellence, dAIEDGE contributed to the event and further strengthened the collaboration between the two projects.

The Networks of Excellence in AI ENFIELD and dAIEDGE address complementary dimensions of next-generation AI. While ENFIELD focuses on adaptive, green, human-centric and trustworthy AI, dAIEDGE advances distributed, efficient and trustworthy AI at the edge. Both projects contribute to connecting research communities and reducing fragmentation within the European AI ecosystem.

Alain Pagani, coordinator of dAIEDGE, participated in the panel “AI Research Funding in Europe: From NoEs to AI Continent Action Plan and Beyond.” Moderated by Ana Solange Leal from INOVA+, the panel also included Cécile Huet, Head of Unit for Robotics and Artificial Intelligence Innovation and Excellence at DG CNECT, Georgios Spathoulas from NTNU and Dimitrios Tzovaras from CERTH.

The discussion reflected on the achievements of the Networks of Excellence and on how their communities, expertise and collaborative structures can continue beyond the end of the individual projects. Particular attention was given to stronger cooperation across countries and initiatives, closer links between research and industrial uptake, and the role of the Networks of Excellence in future European AI strategies.

dAIEDGE also contributed to the panel “Edge AI: Robustness, Continual Learning and Real-World Deployment,” with Ovidiu Vermesan from SINTEF discussing key challenges for adaptive AI systems in real-world edge environments.

The participation of dAIEDGE reflects the close cooperation between the two Networks of Excellence and their shared ambition to strengthen long-term collaboration within Europe’s AI research and innovation community.

Contact: Dr. Alain Pagani

The European research project IRIS-XR – Immersive Realism through Integrated Sensing and Optics for eXtended Reality – held its kick-off meeting on 11 and 12 June 2026. The project develops innovative photonic and AI-based technologies for next-generation extended reality systems, with the aim of enabling more natural, efficient and immersive interaction in virtual and augmented environments.

IRIS-XR combines advanced display technologies with integrated sensing and intelligent perception. Key developments include display-integrated infrared illumination for eye tracking, event-based depth sensing for hand and gesture analysis, and AI-based methods for the joint interpretation of gaze, hand movements and the surrounding environment. The resulting technologies will be integrated into a new XR hardware and software platform designed to improve interaction quality while reducing latency, cognitive effort and energy consumption.

DFKI contributes its expertise in computer vision, machine learning and human-centred XR. The Augmented Vision research department leads work on egocentric hand and gesture analysis, contributes to the integration and orchestration of multimodal sensor data, and is responsible for the technical validation of the developed platform. The technologies will be evaluated in application scenarios covering healthcare, flight simulation and automotive use cases.

Duration: 01.06.2026 – 31.05.2029
Programme: Horizon Europe – Research and Innovation Action (RIA)
Research Department: Augmented Vision
Partners: DFKI, Varjo, VoxelSensors, ITI, Mercedes-Benz, Reiser, University College Dublin, Australo,
Coordinator: Dr. Alain Pagani
Research topics: Computer Vision, Artificial Intelligence, Extended Reality, Human-Computer Interaction, Eye Tracking, Hand and Gesture Analysis

Contact: Dr. Alain Pagani

We are pleased to announce that the paper
“Bounding Box Label Propagation for Re-Annotation of Document Layout Analysis Datasets“
by Nick Jochum, Tobias Alt-Veit, Christian Schön, Alexander Lück, René Schuster, and Didier Stricker
has been accepted for presentation at the 20th International Conference on Document Analysis and Recognition (ICDAR), taking place on August 30th to September 4th, 2026, in Vienna, Austria.

The paper connects the real-world problem of evolving class sets to AI-based document layout analysis.

Special recognition goes to Nick Jochum for carrying out the thesis work that formed the foundation of this publication, as well as to Insiders Technologies for the valuable collaboration.

Congratulations to all authors!

Contact: Dr. René Schuster

We are happy to be part of the Computer Vision Conference, where we have presented our paper “SF3D-RGB: Scene Flow Estimation from Monocular Camera and Sparse LiDAR” by Rajai Alhimdiat, Ramy Battrawy, René Schuster, Didier Stricker, Wesam Ashour.

The paper presents results of our collaboration with the Islamic University of Gaza on sensor fusion for 3D scene flow estimation.
Our work has received the Emerging Research Award, highlighting the quality of the presented paper.

The Computer Vision Conference 2026 took place in Amsterdam, NL, from May 21st to 22nd.

Contact: Dr.-Ing. René Schuster

The Augmented Vision department has 3 papers accepted at the upcoming ICPR 2026 conference.
The conference will be taking place at the International Convention Center, Lyon, France, from August 17th to 22nd. 
The accepted papers are:

PanoSAMic: Panoramic Image Segmentation From Sam Feature Encoding And Dual View Fusion
Mahdi Chamseddine, Didier Stricker, Jason Rambach

MILE: Mixture Of Incremental Lora Experts For Continual Semantic Segmentation Across Domains And Modalities
Shishir Muralidhara, Didier Stricker, René Schuster

Efficient Post-Hoc Calibration In Object Detection Without Held-Out Data
Nikolas Ebert, Didier Stricker, Oliver Wasenmüller

Contact: Shishir Muralidhara

Several researchers of the Augmented Vision department have recently received prestigious reviewer awards from the leading computer vision and machine learning conferences. 

4 AV members (Dr. Jason Rambach, Dr. Alain Pagani, Shaoxiang Wang and Yaxu Xie) have received the Outstanding Reviewer of the CVPR 2026 conference, given to the top 6.1% out of 25.149 reviewers. The IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) is the premier annual computer vision event.  CVPR 2026 will be held in Denver, CO, from June 3^rd to 7^th, 2026. 

The complete list of outstanding reviewers for this year can be accessed HERE 

List of all AV papers to be presented at CVPR. 

In addition, Shashank Mishra has received the Gold Reviewer award of the ICML conference, given to the top 25% reviewers. The International Conference on Machine Learning (ICML) is the premier Machine Learning conference world-wide. ICML 2026 will be held Seoul, South Korea, from July 6^th to 11^th, 2026.    

 
Contact: Dr. Jason Rambach, Dr. Alain Pagani 

Dr. Jason Rambach, served as the main organizer and moderator of the 4th workshop on “AI and Robotics in Construction” at the European Robotics Forum 2026 in Stavanger, Norway (23.3-27.3) supported by the projects of Tech4Construction cluster and the EuRobotics Topic group Construction.

The workshop included presentations on the current state of 6 EU Horizon projects on construction robotics (Beeyonders, RobetArme, ShieldBoT, Target-X, XSCAVE, DISCOVER), followed by a round table and interactive session with the audience on the challenges and opportunities of robotic projects in construction.

Workshop schedule: https://shieldbot.eu/shieldbot-at-the-european-robotics-forum-2026-workshop-on-robotics-and-ai-with-the-tech4euconstruction-projects/

Contact: Dr. Jason Rambach

1st Workshop on Sustainability and Resource-Efficiency of Artificial Intelligence @ IJCAI 2026

Together with the RPTU Kaiserslautern-Landau, the Heidelberg University, and the Korea University, DFKI is organizing the SuRE workshop, co-located with the IJCAI-ECAI 2026 taking place in Bremen, Germany.
The workshop is a part of the Sustainable Embedded AI project and co-sponsored by the Carl Zeiss Foundation.
We would be glad if you consider our workshop for submission of your current work. The Call for Papers features three diverse tracks.

Contact: Dr.-Ing. René Schuster

The department Augmented Vision has presented 3 papers at the WACV 2026 conference taking place on March 6-10, 2026 in Tucson, USA.

WACV, the IEEE/CVF Winter Conference on Applications of Computer Vision, is a top-tier conference for computer vision topics.

The 3 accepted papers are:

IMKD: Intensity-Aware Multi-Level Knowledge Distillation for Camera-Radar Fusion
Shashank Mishra, Karan Sanjay Patil, Didier Stricker, Jason Raphael Rambach

Inpaint360GS: Efficient Object-Aware 3D Inpainting via Gaussian Splatting for 360° Scenes
Shaoxiang Wang, Shihong Zhang, Christen Millerdurai, Rüdiger Westermann, Didier Stricker, Alain Pagani,

TalkingPose: Efficient Face and Gesture Animation with Feedback-guided Diffusion Model
Alireza Javanmardi, Pragati Jaiswal, Tewodros Amberbir Habtegebrial, Christen Millerdurai, Shaoxiang Wang, Alain Pagani, Didier Stricker

Contact: Alain Pagani, Jason Rambach

The department Augmented Vision has 4 accepted papers at the upcoming CVPR 2026 conference taking place from June 3-7, 2026 at the Colorado Convention Center, Denver, USA.

The CVPR conference is the premier international conference in computer vision and pattern recognition.

The 4 accepted papers are:

DriverGaze360: OmniDirectional Driver Attention with Object-Level Guidance

Shreedhar Govil, Didier Stricker, Jason Rambach

LiREC-Net: A Target-Free and Learning-Based Network for LiDAR, RGB, and Event Calibration

Aditya Ranjan Dash, Ramy Battrawy, René Schuster, Didier Stricker

SIMSPINE: A Biomechanics-Aware Simulation Framework for 3D Spine Motion Annotation and Benchmarking

Muhammad Saif Ullah Khan, Didier Stricker

ReLaGS: Relational Language Gaussian Splatting

Yaxu Xie, Abdalla Arafa, Alireza Javanmardi, Christen Millerdurai, Jia Cheng Hu, Shaoxiang Wang, Alain Pagani, Didier Stricker