Authors:
Kamil Kowol
1
;
Stefan Bracke
2
and
Hanno Gottschalk
1
Affiliations:
1
School of Mathematics and Natural Sciences, University of Wuppertal, Gaußstraße 20, Wuppertal, Germany
;
2
Chair of Reliability Engineering and Risk Analytics, University of Wuppertal, Gaußstraße 20, Wuppertal, Germany
Keyword(s):
Corner Case, Human-centered AI, Human-in-the-Loop, Automated Driving, Test Rig.
Abstract:
The overall goal of this work is to enrich training data for automated driving with so-called corner cases in a relatively short period of time. In road traffic, corner cases are critical, rare and unusual situations that challenge the perception by AI algorithms. For this purpose, we present the design of a test rig to generatesynthetic corner cases using a human-in-the-loop approach. For the test rig, a real-time semantic segmentation network is trained and integrated into the driving simulation software CARLA in such a way that a human can drive on the network’s prediction. In addition, a second person gets to see the same scene from the original CARLA output and is supposed to intervene with the help of a second control unit as soon as the semantic driver shows dangerous driving behavior. Interventions potentially indicate poor recognition of a critical scene by the segmentation network and then represent a corner case. In our experiments, we show that targeted and accelerated en
richment of training data with corner cases leads to improvements in pedestrian detection in safety-relevant episodes in road traffic.
(More)