Risk & Planning
Planning complex maneuvers in real environments usually implies taking risks. For humans, these risks are qualitatively related to utility, comfort and social norms based on experiences and observations. In comparison, Artificial Intelligence systems can quantify risks based on a variety of parameters and influences. The result is a range of choices, each one having a calculated risk assigned. Complex planning supported by intelligent systems is therefore able to identify beneficial trajectories in the trade-off between utility and risk under the premise of machine ethics, e.g., for the use in autonomous driving or robot motion.
Risk and planning topics adress questions like:
- How do we move smoothly through a crowded airport hall?
- How do we plan an overtaking maneuver on a busy highway?
A major challenge when estimating risks is the uncertainty related to the prediction of future events due to the unknown future behavior of others.
HRI-EU’s research focuses on quantitative risk models for the evaluation of different prototypical behavior alternatives. A risk map then serves to identify the spots of highest risks, which we use for near-optimal behavior planning.
Driving Support using Predictive Risk Maps
A key attribute of intelligent mobility systems is the capability to smoothly find safe trajectories in highly dynamic environments. This involves the evaluation of own behavior alternatives in the context of the anticipated scene dynamics, e.g., the predicted evolution of the other traffic participants.
The associated risk with a behavior alternative can be represented in so-called “risk-maps”, which quantify risk as a function of prediction time and ego-behavior. Two factors influence the behavior planning: (1) the risk associated with a determined behavior and (2) the resulting utility/gain in terms of mobility goals and comfort constraints.
The optimal behavior is a trade-off between risk, utility and comfort given by a path through the risk map which avoids the peaks of high risk.
Furthermore, the risk map representation can be used as a model to predict the intended behavior of others, and to detect whether they follow a risk-minimizing strategy. Risk estimation technologies are also the basis for future “guardian” type driving assistance systems aiming to enhance safety by recommending and supporting low risk driving strategies. They constitute a core ingredient towards concretizing the vision of zero accident, zero fatality mobility systems.
For more information
T. Puphal, M. Probst and J. Eggert, “Probabilistic Uncertainty-Aware Risk Spot Detector for Naturalistic Driving”, IEEE Transactions on Intelligent Vehicles, 2019