Loading...
 

Image  

Selected Topics: Prediction Technology



Conventional Advanced Driver Assistance Systems (ADAS) merely react to currently sensed situations. Prediction technology researched at HRI allows the reaction to situations even before they can be perceived.

Application Example:

Intelligent Adaptive Cruise Control – i-ACC

In a collaborative project between Honda Automobile R&D Center, Tochigi, Japan; Honda R&D Europe Germany; and HRI-EU such a predictive ADAS was developed – the Honda
i-ACC. This system predicts if a vehicle on a neighboring lane will cut-in and adapts its speed, even before the vehicle started the lane change. Thanks to the prediction technology, i-ACC can react much earlier and smoother to a cutting-in car
than the conventional ACC, thus increasing comfort and safety. Details on the used prediction technology can be found below. This technology is scheduled to debut in the 15.5 European CR-V, offering the world-first Assistance beyond sensing.

See press release.

Portfolio of Prediction Technologies at HRI-EU

To be prepared for the challenges of  future mobility, a portfolio of various prediction methods is researched at HRI. High performance, robust prediction is achieved by a combination of these complementary approaches.

Context-based prediction

Traffic participants usually do not act without reason, but react in the context of their situation. The context-based prediction models and evaluates this context to estimate a traffic participant’s future behavior.

On highways, it evaluates relations between vehicles, in inner-city, it evaluates e.g. relations between pedestrians, vehicles, and traffic elements (see image).






Context-based prediction is well suited
for early prediction.

Physical prediction

Information from automotive sensors needs to integrated to allow accurate spatial prediction in unstructured environments.

This prediction spans a spatial grid of cells and computes a probability that a traffic participant is at a certain position at a certain point in time.






Physical prediction is very robust and
not dependent on context information.

Bayesian grid based prediction

Information from automotive sensors needs to integrated to allow accurate spatial prediction in unstructured environments.

This prediction spans a spatial grid of cells and computes a probability that a traffic participant is at a certain position at a certain point in time.





Bayesian grid based prediction is capable

of handling noise and different process models.


See publication on this topic.

Spatio-temporal prediction

The behavior of traffic participants depends on the future spatial arrangement of them and their environment.

Spatio-temporal prediction estimates future arrangements between traffic participants over time. Based on this prediction it evaluates the optimal ego vehicle behavior.






Spatio-temporal prediction is valuable especially for navigating in complex situations.

Risk prediction

Conventional prediction methods are effective for foreseeing the most likely situations. However, prevention of accidents requires prediction of rare but dangerous situations.

Risk prediction approaches this problem by estimating the probability of damage in a certain situation with the probability of reaching this situation from the current situation. Varying over different situations results in a risk map which is used to find the risk-minimizing behavior.






Risk prediction is capable of handling

rare but severe situations.


See publication on this topic.