Simon Hecker , Dengxin Dai and Luc Van Gool
ETH ZurichAutonomous vehicles are more likely to be accepted if they drive accurately, comfortably, but also similar to how human drivers would. This is especially true when autonomous and human-driven vehicles need to share the same road. The main research focus thus far, however, is still on improving driving accuracy only. This paper formalizes the three concerns with the aim of accurate, comfortable and human-like driving. Three contributions are made in this paper. First, numerical map data from HERE Technologies are employed for more accurate driving; a set of map features -- which are believed to be relevant to driving -- are engineered to navigate better. Second, the learning procedure is improved from a pointwise prediction to a sequence-based prediction and passengers' comfort measures are embedded into the learning algorithm. Finally, we take advantage of the advances in adversary learning to learn human-like driving; specifically, the standard L1 or L2 loss is augmented by an adversary loss which is based on a discriminator trained to distinguish between human driving and machine driving. Our model is trained and evaluated on the Drive360 dataset, which features 60 hours and 3000 km of real-world driving data. Extensive experiments show that our driving model is more accurate, more comfortable and behaves more like a human driver than previous methods.
Paper / BibTeX / Dataset (released) / Workshop
This dataset is released as part of our ICCV 2019 Workshop "Learning to Drive" Challenge. Please register for the competition to get the dataset which will provide a download link once registered. Challenge particpants, upon signing up, will receive 1) three csv files (one for each of the three sub-sets) that specify the synchronized image paths, road attributes, GPS and the CAN bus control labels (for the train and validation set); and 2) a link to all the images extracted from the videos.
An illustration of HERE map features used in this work. Please find the details in the paper.
Our method is more accurate, more comfortable and behaves more like a human driver than previous methods. See the paper for more results and more details.
Acknowledgement
The work is supported by Toyota Motor Europe via the project TRACE-Zurich. We would like to thank HERE Technologies, for granting us the access of their map data, without the support this project would not have been possible.
