Deadlock detection, cooperative avoidance and recovery protocol for mixed autonomous vehicles in unstructured environment

[摘要] Deadlock is an extreme traffic flow operational state during rush hours. Many literatures have studied autonomous vehicle coordination under the umbrella of deadlock-free conditions. These researches either assume the trajectories are fixed or state spaces are discrete and limited on structured road spaces or don't consider the influence of human-driven vehicles (HDV), which are not controllable from the system's viewpoint. This manuscript relaxes the above limitations and proposes a method to detect, avoid, and recover from deadlock for mixed autonomous vehicles flow. Firstly, two types of deadlocks, weak and strong , are defined based on deadlock properties. Next, two detection algorithms based on evasion distance propagation are proposed. After that, we present a cooperative control method to avoid deadlock based on chain-spillover-free and loop-free strategies. If a deadlock has already happened, cooperative protocols based on re-routing and backward-forward strategies are designed. The proposed model is tested in Carla. The results show that the deadlocks can be detected 13 seconds earlier than their occurrence, and it takes about 6 seconds to unlock the existing deadlock. The results also show that with the proposed deadlock avoidance algorithm, the traffic throughput can be increased by 35.7%, and with the proposed deadlock recovery protocol, the traffic throughput can be increased by another 18%.

[效力级别]  Early Access

[关键词] STOCHASTIC PETRI-NET;MODEL