On the Tradeoff between Hardware Protection and Optimization Success: A Case Study in Onboard Evolutionary Robotics for Autonomous Parallel Parking

Making the transition from simulation to reality in evolutionary robotics is known to be challenging. What is known as the reality gap, summa-rizes the set of problems that arises when robot controllers have been evolved in simulation and then are transferred to the real robot. In this paper we study an additional problem that is beyond the reality gap. In simulations, the robot needs no protection against damage, while on the real robot that is essential to stay cost-effective. We investigate how the probability of collisions can be minimized by introducing appropri-ate penalties to the fitness function. A change to the fitness function, however, changes the evolutionary dynamics and can influence the opti-mization success negatively. Therefore, we detect a tradeoff between a required hardware protection and a reduced efficiency of the evolutionary optimization process. We study this tradeoff on the basis of a robotics case study in autonomous parallel parking.