Do you want to work on The vehicle safety envelope?

Objectives

1. The EV comes with quite some benefits w.r.t environment, lower to zero carbon emissions and zero
   urban air pollution. The objective is to eliminate negative impact on traffic safety, due to the low
   noise emissions, causing less perception of approaching vehicles by other vulnerable road users,
   like elderly, pedestrians and/or cyclists.
2. To address the challenges from 1) develop and introduce a safety shell around the vehicle.
3. Obtain a good understanding of an autonomous safety shell around a car, which autonomously
   maximally prevents accidents or minimizes accident-impact by active anticipating actuation within
   the limits of the vehicles capabilities, and then continues-gives back control to the driver.
4. Give insight in the ways the safety shell may activate automated, advisory or warning signals to
   the EV-driver or vulnerable other road users, by auditive, tactile, visible, or vehicle control
   measures.

Activities

1. Sensor data interpretation and fusion techniques for anticipatory safety systems.
   a. Safe driving requires anticipating on possible future events. Deep learning has, so far, solved
   object detection and classification. The next step is to automatically hypothesize what these
   objects, e.g. vehicles and vulnerable road users, will be doing in the future. In complex safety
   scenarios, this requires a holistic scene understanding approach that is able to detect and
   interpret (inter-) object behavioral cues embedded deep in the spatio-temporal domain.
   b. This research line focusses on automatically learning these behavioral cues from big data using
   semi-supervised techniques. The innovation is in the training strategy that takes as input many
   positive examples, i.e. safe situations, and only few to none negative examples, i.e. (near)
   collision situations.
   c. The solution that will be explored is casting this problem into a regression task, where a deep
   Generative Adversarial Network (GAN) is trained to i) regress on positive examples, and ii)
   learned to generate and extrapolate negative samples.
   d. The trained GAN is embedded in the safety shell system, to feed world modeling and motion
   planning.
2. World Modeling and Motion planning for EV Safety Shell
3. To define a safety-shell for an autonomous vehicle, understanding and modeling the
   context/situation the car is in is crucial.
   a. Looking into classes of scenarios and complex situations rather than single use case (e.g. define
   the safety shell with respect to. various Road Users)
   b. The search of the free space for maneuvering and for fail-safe situations
4. Determine regions of interest or regions of care (i.e. safety shell), for all possible maneuvers and
   incorporate these into motion planning.
   a. Investigate with respect to what are these regions parametrized or if they change with context.
   b. Which path can we plan using the information of the safety shell, such that no or nearly 0
   collisions exit.

Expected output

1. At least 6 journal papers describing the results of the safe mobility research (M12x1, M24x1,
   M36x2, M48x6).
2. At least 4 conference papers. Most co-authored with other WPs (M24x2,M48x2).
3. A tested concept system in 1 digital twin in 1 of the 5 representative neighborhoods in WP10
4. At least 2 workshops in which the safety shell concept is explored with NEON stakeholders
   interactively.
5. At least 1 PhDs will have spent more than 3 months abroad with knowledge institutions or from
   abroad hosted by NEON.
6. At least 5 mentions of Safety shell concept in national newspapers or on television.