Scope and Topics
Today, most automated driving systems still heavily depend on HD maps.
They serve as a crucial information source for various safety-critical tasks such as localization, prediction and behavior planning.
By not suffering from limited range, affected by occlusions or dependent on favorable weather/light conditions, HD maps extend the foresight of an autonomous system and the robustness towards sensor failure.
At the same time HD maps are expensive to acquire and even more expensive to maintain due to frequent changes of the environment.
Even with huge efforts in keeping HD maps updated, they stay a snapshot of the past and can obtain outdated information, while being a crucial requirement for map-based driving.
In light of this situation, many researchers tackle the future vision of map-less driving:
The recent developments in sensor, sensor fusion and machine learning technologies enable a rapidly improving online modeling of the vehicle’s surrounding and on-the-fly estimation of HD map information.
While there are already many approaches to predict the surrounding semantics and detect/track objects, current research also addresses the online estimation of higher-level scene understanding layers such as lane topology and traffic rules.
While both philosophies have their strengths and are frequently considered on their own, they can both benefit from the unique advantages of the respective other.
This workshop aims to bring both perspectives closer together and explore how future approaches might consider a hybrid solution. However, the workshop is not limited to this.
It welcomes a wide range of contributions regarding map-based and map-less driving solutions.
The topics of interest of the workshop include, but are not limited to:
- Online fusion of uncertainty-affected on-the-fly map estimation with potentially outdated HD maps.
- Motion and behavior planning considering uncertainty-affected map estimation and/or potentially outdated maps.
- Certifiable robustness / validation of map-based, map-less or hybrid systems.
- Data sets for learning higher-level scene understanding e.g. lane topologies, traffic rules.
- Leveraging existing HD maps to generate training data for on-the-fly road modeling and map estimation.
- Online road modelling and on-the-fly map estimation for autonomous driving..
- Map change detection, map update and map validation.
