Page - (000110) - in Autonomes Fahren - Technische, rechtliche und gesellschaftliche Aspekte

Implementable Ethics for Autonomous Vehicles98 Humans tend to accept or, in some cases, expect these sorts of actions from other humans. Drivers who drive at the speed limit in the left hand lane of a highway may receive indica- tions, subtle or otherwise, from their fellow drivers that this is not the expected behavior. But will these same expectations translate to automated vehicles? The thought of a robotic vehicle being programmed to systematically ignore or bend traffic laws is somewhat unsettling. Yet Google’s self-driving cars, for instance, have been programmed to exceed the posted speed limit on roads when commanded by the operator [20]. Furthermore, there is little chance that the driver annoyed by being stuck behind another car traveling the speed limit in the left lane of the freeway will temper that annoyance because the car is driving itself. Our current expectations of traffic flow and travel time are based upon a somewhat fluid application of traffic laws. Should automated vehicles adopt a more rigid interpreta- tion and, as a consequence, reduce the flow or efficiency of traffic, societal acceptance of these vehicles might very well suffer. If automated vehicles are to co-exist with human drivers in traffic and behave similarly, a deontological approach to collision avoidance and a consequentialist approach to the rules of the road may achieve this. 5.6 Simple Implementations of Ethical Rules Some simple examples can easily illustrate the consequences of treating ethical goals or traffic laws as rules or costs and the different behavior that can arise from different weights on priorities. The results that follow are not merely drawings but are rather simulations of algorithms that can be (and have been) implemented on automated vehicles. The exact mathematical formulations are not included here but follow the approach taken by Erlien et al. [21, 22] for collision avoidance and vehicle automation. These references provide details on the optimization algorithms and results of experiments showing implementation on actual test vehicles. To see the interaction of costs and constraints in vehicle decision-making, consider a simple case of a vehicle traveling on a two lane road with an additional shoulder next to the lanes (Figure 5.4). The goal of the vehicle is to travel straight down the center of the given lane while steering smoothly, using the cost function for path tracking and steering from Equation 5.2. In the absence of any obstacles, the car simply travels at the desired speed down its lane and none of the constraints on the problem are active. Fig. 5.4 The basic driving scenario for the simulations. The car is traveling on a straight two-lane road with a shoulder on the right and approaches an obstacle blocking the lane