Conceitos essenciais
A novel deep reinforcement learning algorithm that effectively considers the behavior of both leading and following vehicles to enhance longitudinal control and collision avoidance in high-risk driving scenarios.
Resumo
This study introduces a deep reinforcement learning-based algorithm for longitudinal control and collision avoidance in advanced driver assistance systems (ADAS). Existing ADAS technologies, such as adaptive cruise control (ACC) and automatic emergency braking (AEB), primarily focus on the vehicle directly ahead, often overlooking potential risks from following vehicles. This oversight can lead to ineffective handling of high-risk situations, such as high-speed, closely-spaced, multi-vehicle scenarios where emergency braking by one vehicle might trigger a pile-up collision.
To overcome these limitations, the proposed algorithm effectively considers the behavior of both leading and following vehicles. The study utilizes the Deep Deterministic Policy Gradient (DDPG) reinforcement learning model to navigate complex vehicle-following situations and accommodate various vehicle types with different acceleration policies.
The algorithm was evaluated in simulated high-risk scenarios, including emergency braking in dense traffic and multi-vehicle following scenarios. The results demonstrate the algorithm's ability to prevent potential pile-up collisions, including those involving heavy-duty vehicles, which traditional ADAS systems typically fail to address.
The key highlights of the study include:
Development of a vehicle brake and acceleration policy that enhances safety by addressing the potential safety risks from the following vehicles through the exploration of edge case collision scenarios.
Development of a universally applicable algorithm designed to mitigate the incidence of serious pile-up collisions.
Simulation studies showing that the DDPG-based algorithm effectively reduced collisions that traditional methods cannot avoid.
Estatísticas
The leading vehicle activates emergency braking at a deceleration of -3 m/s^2.
The heavy following vehicle exhibits a lower maximum deceleration of -6 m/s^2 compared to the light following vehicle with a standard AEB deceleration of -7.5 m/s^2.
Citações
"The proposed algorithm has the capability to dynamically select different deceleration in response to the behavior of both leading and following vehicles."
"The RL algorithm optimally calculates the deceleration at each time step, allowing the ego vehicle to stop within the gap without any collisions."
"The vehicles in the middle, which are controlled by the proposed RL algorithm, exhibit dynamically changing responses in terms of deceleration and acceleration."