The safety of Autonomous Vehicles (AVs) remains a key challenge for developers and stakeholders due to the dynamic environments in which they operate. From a control perspective, nonlinear model predictive control (MPC) is a powerful control strategy to operate in such dynamic environments. This paper evaluates the performance and compares the safety criteria for AV controllers designed using MPC and control barrier functions (CBFs) that ensure safety in obstacle avoidance through the principle of set invariance. With CBFs, there are two design approaches: CBFs formulated as a discrete-time constraint (MPC-CBF) or as a quadratic program (MPC-CBF-QP). In addition to CBFs, a more straightforward approach for obstacle avoidance is to use the safe-distance constraints in the MPC formulation (MPC-DC). The results of these three nonlinear MPC control strategies: (I) MPC-DC, (II) MPC-CBF-QP, and (III) MPC-CBF, implemented on an actual vehicle in autonomous mode, are discussed in detail. Their performance and safety conditions are compared in three common urban and highway driving scenarios: (i) avoiding static obstacles, (ii) sudden pedestrian interaction, and (iii) overtaking the lead vehicle (LV). We experimentally and theoretically show that the MPC-CBF-QP is computationally efficient and guarantees safety in all considered scenarios. MPC-DC often fails to meet safety requirements in critical scenarios, and MPC-CBF performance heavily depends on prediction, control, and CBF horizon.
扫码关注我们
求助内容:
应助结果提醒方式:
