Environmental model extension for lane change prediction with neural networks

Autoren

Martin Krüger, Anne Stockem Novo, Till Nattermann, Manoj Mohamed, ZF Group, Automated Driving & Integral Cognitive Safety;
Torsten Bertram, TU Dortmund, Fakultät für Elektrotechnik und Informationstechnik

Jahr

2019

Zusammenfassung

Driven by the desire of improving traffic safety, traffic efficiency and a better utilization of the time people spend in traffic, the development proceeds from Advanced Driver Assistance Systems (ADAS) towards fully automated systems. ADAS systems are categorized as level 2 systems according to the Society of Automotive Engineers (SAE) definition, on a classification scheme from level 0 to 5 where level 5 is a fully automated system. With increasing automation level the complexity rises, due to the piecewise shift of responsibility towards the automated driving system.
In accordance with Tas et al. any structure of an automated driving system is composed of three basic blocks, detection and perception, situation analysis and motion planning, and control. First, an environmental sensor system, that may be composed of different sensor types perceives the surrounding of the automated vehicle and fuses the information to build an environmental model. The subsequent planning module further processes the information of the environmental model to evaluate the surrounding traffic participants behavior and to plan the motion of the ego vehicle (EV). Finally, the motion that was planned for the EV is executed and controlled.
For the motion planning of the EV, the evaluation of all the surrounding vehicles is important. To plan the motion of the EV on a trajectory and a strategical level, an estimation of the future state of the impending vehicles is of special interest. Since any vehicle is particularly influenced by vehicles on its ego lane and the neighboring lanes, the three closest impending vehicles on these lanes are considered for this study. For these three vehicles mentioned above, the current and future driving maneuver is determined.
The maneuvers that are distinguished within this work are lane keeping (LK), lane change left (LCL) and lane change right (LCR). Since a lane change (LC) maneuver is defined as the entire process when the target vehicle (TV) moves all the way from its initial position until the actual crossing of the lane marking, the early detection of a LC maneuver means predicting the LC in advance.

[...]

The remainder of this paper is structured as follows. In section 2 related works are discussed. Section 3 introduces the proposed new features. Results according to these features and further investigations are presented and discussed in section 4. Section 5 concludes the paper.

Mitglieder-Login