Information fusion for secure self-driving cars

Even though the self-driving revolution is banging on the door, there still is a safety problem. Besideds, current self-driving technologies like LIDAR, is expensive. The EIT Digital Doctoral School now offers a Industrial Doctorate programme, run by Bosch Hungary, to improve autonomous systems by fusing different information sources. The title of the programme is Information fusion of video analysis in ADAS system.

Most self-driving cars use the LIDAR technology. LIDAR stands for Light Detection and Ranging. With Lidar self-driving cars can observe, via firing off beams of laser light, the world with a continuous 360 degrees of visibility and accurate depth information, like measuring the distance between an object and you.

But, LIDAR-based technology remains expensive and is unlikely to result in affordable cars for the mass market. Alternative solutions based on Artificial Intelligence (AI) would be more affordable but despite the considerable advances in machine vision propelled by the advent of deep networks they are not providing absolute safety. A single deep algorithm cannot be trusted as it can misinterpret potential dangers and cause accidents. The strength of fusing different information sources, rules, constraints, and physical models should be studied to overcome this issue. EIT Digital Doctoral School’s PhD programme will require exactly that kind of activity.


Tesla’s accident, where the car on autopilot hit the trailer of a truck after it didn’t notice the vehicle against a brightly lit sky, perfectly shows that minor aspects of the visual information may change and spoil the observations. Handling unusual, unexpected situations is one of the most difficult tasks. One major issue is facing false positive that result in undesired counter measures or in delayed decisions.

Some works at the Neural Information Processing Group on Cognition, spatio-temporal correlations using deep learning and on human-AI interaction and 2D-to-3D mapping describe the problem and offer a solution that also includes anomaly or salience detection in a joined deep learning and traditional AI architecture capable of self-training. Such issues are relevant for Bosch Hungary.


Among the methods will be the use of database for improved representation of situations such as 2D-to-3D mapping and temporal prediction, re-evaluation of past representations given novel observations, maintaining a single and consistent interpretation, predicting and learning from prediction. These are all “in progress”. Furthermore, you will need to adapt anomaly detection using both learned models, unresolved interpretations, and outliers in the predictions.

In addition, exploiting state of the art deep-learning methods, such as object detection, semantic segmentation will be required. Other miscellaneous activities include posing estimation, monocular depth estimation, optical flow, boundary flow as well as structuring from motion, saliency objectness, and vanishing point detection.

Another feature of the PhD work will be using databases, such as KITTI, Cityscapes, Mapillary, MS COCO plus databases that will be released in the near feature. It is worth to note that there is a doctoral activity with Mapillary that can create data for this proposed one, and synergies are evident.

Expected outcome

The expected results of the PhD is a self-improving model that uses state of the art algorithms and gives rise to complex and reliable description of the environment in space and time and has value for self-driving cars. Detailed analysis of the used algorithms is also expected. On the other hand, prototype of the new model for the designers and engineers for thorough testing and hardware development that will be acquired by Bosch. The PhD student should publish papers in high-quality academic conferences and journals. Consulting with Bosch throughout the doctoral work on the embedding of the methods into self-driving cars under observational and computational time constraints ensures the alignment with the industrial constraints and ease final technology transfer.


The doctoral student involved in this industrial doctorate programme will share his time between the EIT Digital Doctoral Training Centre in Budapest, the premises of Bosch and/or ELTE Soft, and ELTE.


  • Industrial partner: Bosch and/or ELTE Soft
  • Academic/research partner: ELTE
  • Number of available PhD positions: 1
  • Duration: 4 years
  • This PhD will be funded by EIT Digital, ELTE, and Bosch.


If you are interested in applying for this position, please send an e-mail to the EIT Digital Budapest DTC lead, including a CV, a motivation letter, and documents showing your academic track records.

Please apply before 14 September 2018, 12.00 CET.

© 2010-2020 EIT Digital IVZW. All rights reserved. Legal notice