Learning scenario-specific vehicle motion models for intelligent infrastructure applications
| Autor: | Christian-Eike Framing, Frank-Josef Heßeler, Dirk Abel |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
0209 industrial biotechnology
Digital mapping Computer science 020208 electrical & electronic engineering Real-time computing 02 engineering and technology Motion (physics) 020901 industrial engineering & automation Side effect (computer science) Control and Systems Engineering 0202 electrical engineering electronic engineering information engineering Trajectory Probability distribution Scale (map) Intersection (aeronautics) |
| Zdroj: | 7 Seiten (2019). doi:10.18154/RWTH-2019-04971 |
| ISSN: | 2405-8963 |
| DOI: | 10.1016/j.ifacol.2019.08.057 |
| Popis: | The development of autonomous driving promises to make traffic safer, more comfortable and efficient. While respective vehicles are equipped with an array of sensors, communication and intelligent infrastructure systems will act in a supporting role for perceiving and predicting the surrounding traffic environment. As a side effect it becomes possible to collect vehicle trajectories on a large scale and deduce vehicle motion behavior specific to a traffic scenario of interest. In this work we present our LSTM-MDL model that learns vehicle motion behavior entirely from collected trajectory data and does not rely on previously available lane-accurate digital maps. It is evaluated on a real-world dataset from an innercity intersection. The resulting prediction output consists of a probability distribution of future vehicle trajectories that is multi-modal where multiple driving maneuvers are possible. Trajectory prediction is shown to be conditional on the used lane and velocity of a vehicle. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|