What Does a Good Prediction Look Like?

In the developement of safe intelligent vehicles, intentions of other road users need to be estimated in order to plan a safe and convenient trajectory. The present work tries to approach the answer to the question of how a prediction could be assessed holistically using the example of pedestrian prediction. Recently used metrics are compared on a state of the art (SOTA) prediction method. A good prediction does not only have to verify itself on a limited data set, so machine versus human pedestrian prediction are compared. A study with test persons points out the similarity of human predictions compared to SOTA prediction. However, further research is required since we cannot propose a holistic metric that answers the question in the title, yet.