Maneuver prediction using vehicle sensor data

  • Sponsored by: BMW
  • Project Lead: Dr. Ricardo Acevedo Cabra
  • Scientific Lead: Nico Epple 
  • Term: Summer semester 2019

Results of this project are explained in detail in the final documentation and presentation.

With the demand of developing highly automated driving, the number of advanced driver assistance systems (ADAS) is rising and new functions are added to the range of ADAS that support the driver. These functions also warn the driver and intervene in potentially dangerous situations (e.g. braking assistant, lane change assistant). Those situations are of interest for evaluating existing ADAS in the context of an effective analysis. To collect real data of these situations, the BMW Group is conducting naturalistic driving studies (NDS) in different countries.

The goal of this project is to analyze vehicle signals in order to assess and predict different driving maneuvers (e.g. lane change) with respect to different aspects such as kinematics, environment, surrounding objects, route type, individual vehicle guidance or ADAS.

With special focus on lane change maneuvers, lane change detection (auto labeling) and prediction for the observed vehicle should be implemented. Based on identified awaited or aborted events, the influence of lateral guidance systems can be evaluated.

Motivated by the size of data, the NDS is stored in a distributed environment (Hadoop) and processed and finally analyzed in a spark framework. The final machine learning framework for this application could be performed on a different platform.