A Real-time Safe Planning and Control Architecture for Autonomous Driving Adapting to Slippery Roads

Abstract

With autonomous driving being tested – and soon deployed – in more and more challenging environment, it is very important to get guaranties from sub-systems. Although the complexity of the real world makes the elaboration of such guarantees sophisticated, it is still possible to ensure consistency and adaptability of the planning and control part. This means that, at least, the intention and the action undertaken will be executed correctly. This paper presents a real-time safe planning and control architecture that fully adapts to slippery roads. It is based on a 10Hz Model Predictive Control motion planner and 100Hz low-level controllers. Its validity is demonstrated both theoretically and in simulation, by performing tests using a high-fidelity vehicle model on a challenging track with static obstacles on both wet and snow-covered roads. One interesting feature is that, while adapting to extreme low adherence, the algorithms and parameters for planning and control remain the same, showing the capacity of the architecture to adapt to the situation.