BASELABS Data Fusion Forum

Data fusion experts meet practitioners

Our vision

Data fusion as part of the environmental model is one of the key components for automated driving. The performance, reliability and flexibility of this software component will have a decisive influence on the success of vehicle automation. As active members of the community, we initiate the BASELABS Data Fusion Forum.

The event puts an emphasis on interactive formats such as moderated group discussions, extended Q&A sessions or workshops to facilitate learning effects, exchange and networking. We provide a dedicated opportunity with a very limited number of participants to foster in-depth discussions. Based on the principle to offer an event that we ourselves would be excited to join, we defined the following restrictions:

  • Not more than 30 participants to allow in-depth discussions and networking
  • Audience with relevant background (BASELABS reserves the right to select participants based on their practical experience and background in data fusion and perception)
  • Focus on knowledge building and sharing, no sponsored contents
  • Clear focus of topics with an emphasis on technical depth

Event review

In June 2019 a selected group of international data fusion experts from OEMs, Tier1 and others met in Munich for the first edition of the event and discussed topics like low-level fusion, functional safety, validation, and extended object tracking. Our vision has become a reality!

Find the program review and speaker profiles in our event brochure.

Topics of the interactive sessions:

In the interactive sessions, moderated by BASELABS experts, the participants discussed current topics of data fusion and jointly worked out results.

  • Extended objects tracking: Eric Richter, Director Customer Relations
  • Reference data for data fusion evaluation: Norman Mattern, Director Product Development
  • Paradigms for low-level fusion: Christian Adam, Team leader for the BASELABS Create Embedded framework
  • Data fusion and functional safety: Gabi Escuela, Functional Safety Manager
  • Development tools / phases: Marcus Obst, Director Customer Projects
Loading the player ...

Feedback of the participants

We would like to thank the participants for their positive feedback and the many suggestions and ideas for improvement. All participants who completed the feedback form indicated that they would like to participate next year again!

Recommendations:

"All the presentations were very inspiring." Patricio Valenzuela, Scania CV AB

"Great gathering of fusion community with open and interesting discussion." Christian Larsson, Einride AB

"Great networking event. Nice and friendly atmosphere. Charismatic and friendly hosts." Christof Chlebek, Robert Bosch GmbH

"Great place to share experiences with peers that speak your "professional language"." Christian Gießen, Continental AG

"I really appreciate the effort put into the event and also the little details." Gergely Szirmai, AImotive Ltd.

BASELABS Data Fusion Forum 2020

To stay tuned about the event details for 2020, please subscribe to our newsletter. We are looking forward to welcoming you in our data fusion community. The success of the event crucially depends on excellent speakers. While some of the speakers of 2019 already volunteered to serve as a speaker next year (again or for the first time), we will certainly be looking for additional speakers. So, if you know capable people that you would like to see at our event, please contact us.

Impressions Interactive Sessions 2019

In the session "Extended object tracking" we discussed the current challenges and solutions for the handling of high-resolution sensors in the field of object fusion. On the one hand, there was consensus on the necessity of production-ready methods, on the other hand, the excessive complexity of some current methods based on e.g. Random Finite Set (RFS) was critically evaluated. Intermediate status: "Extended object tracking" is needed for higher-value driving functions, solutions ready for series production still have to be developed.

Eric Richter – Data Fusion Expert

“In the session “Reference Data for Data Fusion Evaluation” we discussed different related aspects. Reasons for reference data often being not available were – besides others – seen in the long-lasting and expensive generation process. Also, frequent changes in the sensors and their firmware lead to outdated data sets. Regarding the required quality of reference data, the most important requirement has been seen in an as-low-as-possible false alarm rate. Concerning the metrics calculated by means of the reference data, it was common sense that both function-dependent and function-independent metrics are necessary in the course of the development of data fusion systems.”

Norman Mattern – Product Manager Data Fusion Software

“In the session "Paradigms for Low-Level Fusion" we discussed about fusion of sensor data, which is less pre-processed than with today's systems. These are, for instance, lidar point clouds, radar images, and raw or semantically segmented camera images. Dynamic grid fusion and deep learning have been identified as possible algorithmic approaches. However, there are still many questions to be answered. Can the learned networks be transferred to other sensor configurations? How can the procedures be made robust against incorrect calibration and movements of the body? What does the representation of the environment look like at all? Furthermore, it was discussed whether the required computing power and bandwidth will be available in vehicles and whether sensor manufacturers will open in future to providing raw data at their interfaces. The final aim of low-level fusion is to make more information from the sensor data usable and thus to increase the quality and safety of the system.”

Christian Adam – Low Level Data Fusion Expert

“In the session „Development tools/phases” we discussed the needs and the availability of development tools for data fusion applications in the automotive domain. During the discussion we identified two separate dimensions; 1) the development phase in terms of maturity and 2) the level of tool support.  The development phase was further divided into the prototyping/pre-development and series production phase. Depending on the strategy of a company, there are different models how knowledge and artifacts are transferred between these phases. A classical approach does not implement any direct interaction between the two units of a company. Hence, only requirements and architectural proposals are exchanged. Other companies aim to re-use prototypes and apply hardening phases to reach series production quality.
On a tool level, middleware software and domain specific data fusion tools and SDKs have been discussed. We concluded that a clear separation between both levels is often hard to achieve. For example, proper handling of time stamps and techniques for OOSM processing can be implemented either in the middleware or inside of the data fusion.“

Marcus Obst – Manager Of Data Fusion Series Production Projects

“In the session “Data fusion and functional safety” we discussed about current challenges and approaches related to the development of data fusion systems in the context of critical driving functions. Both sides of functional safety were considered: process-related issues (e.g. finding an efficient way to meet the requirements of ISO 26262) and technical aspects (e.g. centralized and decentralized safety architectures, faults to be considered when analyzing hazards for data fusion processing, dealing with the inherent algorithmic complexity associated with such systems). Regarding the technical aspects, a common opinion was that more regulations and standards are needed to support the development of safe data fusion systems. Moreover, when discussing about the difficulty to specify "good” safety and functional requirements for a data fusion system, it was concluded that there should be no hard transition between prototype (concept) phase and series development, but rather a feedback loop between the requirements, simulation of appropriate scenarios (especially corner cases) and analysis of KPIs that measure the results of the data fusion.”

Gabi Escuela – Functional Safety Manager

Contact and further information

Top of page
This website uses cookies to ensure you get the best experience on our website. By visiting www.baselabs.de you agree to the use of cookies.