The main challenge for autonomous driving is to ensure an adequate level of safety. Considering the almost infinite variability of possible scenarios that autonomous vehicles would have to face, the use of simulators is becoming of utmost importance. They allow the use automated validation and testing in any simulated scenario. It can be used for development for methods relying on closed-loop evaluation, such as machine learning and reinforcement learning approaches. In principle, simulation can be used with all the functional domains of autonomous driving, i.e. route generation, perception, decision making, path planning, and control. However, this powerful methodology is only applicable when the reality gap is low. Modeling this reality gap requires new metrics, which depends on the specific task. In this paper we propose a new approach to assess the real-to-sim gap for a path tracking systems. A real-time model predictive controller (MPC) based on adaptive potential fields was developed and validated using the CARLA simulator. The path-planning and control systems were then tested in real traffic conditions. The gap was evaluated using two different correlation metrics: the Pearson correlation coefficient (PCC) and the max normalized cross-correlation (MNCC). These variables allow optimizing the controller by using the sim-to-real transfer. The controller was evaluated using equivalent simulated and real traffic scenarios. ; JRC.T.3 - Algorithmic Transparency
Sim-to-real transfer and reality gap modeling in model predictive control for autonomous driving
2023-01-01
Miscellaneous
Electronic Resource
English
DDC: | 629 |
Real-Time Implementation of Randomized Model Predictive Control for Autonomous Driving
IEEE | 2022
|Model Predictive Control for Safe Autonomous Driving Applications
Springer Verlag | 2022
|Real-Time Model Predictive Safety Assessment for Level 3 Autonomous Driving
Springer Verlag | 2022
|Trajectory generation for autonomous highway driving using model predictive control
BASE | 2017
|