REHEARSE: adveRse wEatHEr datAset for sensoRy noiSe modEls

Simulation has become an important part of automated vehicle development, but to have an accurate simulation, it is necessary to have correct and accurate sensor models, with this in mind in this project a dataset is created.

The project ROADVIEW uses the V Methodology to implements its algorithms, this implies that to develop and validate algorithms in this project it is required to do simulation. For a simulation to be close to its reality counterpart, the proposed methodology is to use XIL (everything in the loop), but for the XIL environment to close the simulation-to-reality gap, one of the requirements is to have the sensor’s noise model. The importance of closing the simulation-to-reality gap is noted by authorities also such as seen in [1], where the German Ministry has underlined the importance of testing algorithms using simulation and that the simulation-to-reality gap must be small. An example of a technic to lower this gap is seen in [2] where a camera is seen in the simulation loop. Simulation provides the ability to propose safety critical scenarios, and quick down time to change scenarios.

The ROADVIEW project tackles this requirement of a low simulation-to-reality gap in many ways, one of them is by generating sensor noise models. The creation of this noise models requires a dataset with data from the sensors that will be utilized in the XIL environment. As the goal of ROADVIEW is to create a robust adverse weather algorithm, the models must be created to adverse weather conditions. Therefore, this work discusses creation of a dataset that will support the development of the further sensor models. The dataset requires different distances (from close to far), and different weather conditions (ranging from clear to harsh), in such case, the data collection for this work was made in the CARISSMA and CEREMA PAVIN proving Ground, as they have complementary abilities regarding synthetic rain generation, and CE has fog generation. As shown in the below Figure 1, CE (light blue in the figure) can produce more rain but at closer distances compared to CARISSMA (dark blue in the figure), where there is less rain but has a larger total size of the test track.