Deep Reinforcement Learning for Advanced Longitudinal Control and Collision Avoidance in High-Risk Driving Scenarios

TL;DR

The paper tackles safety gaps in traditional ACC/AEB systems that overlook following vehicles in high-risk, dense traffic. It proposes a DRL-based longitudinal control framework using Deep Deterministic Policy Gradient (DDPG) within an MDPlike formulation to account for both leading and following vehicle behaviors, with an 8-dimensional state, continuous action, and a collision-focused reward $r$ (15 for no collision, -3000 for collision) and objective $E\big[\sum_{t=0}^{\infty} \gamma^t r(s_t,s_{t+1},a_t)\big]$. Training with exploration/exploitation over edge-case scenarios demonstrates convergence after ~400 episodes and robust collision avoidance beyond a baseline ACC/AEB, including heavy-vehicle contexts. The results indicate that the RL approach can adaptively modulate deceleration, trigger early follower AEBs, and coordinate multiple vehicles to avert pile-ups, suggesting meaningful safety improvements for high-risk driving in real-world ADAS deployments.

Abstract

Existing Advanced Driver Assistance Systems 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, this study introduces a novel deep reinforcement learning based algorithm for longitudinal control and collision avoidance. This proposed algorithm effectively considers the behavior of both leading and following vehicles. Its implementation in simulated high risk scenarios, which involve emergency braking in dense traffic where traditional systems typically fail, has demonstrated the algorithm ability to prevent potential pile up collisions, including those involving heavy duty vehicles.

Figures (5)

• Figure 1: Model framework
• Figure 2: Potential dangerous collision scenarios
• Figure 3: Change of the Reward during RL training
• Figure 4: Baseline ADAS algorithm implementation in proposed edge case scenarios
• Figure 5: Proposed DRL algorithm implementation in proposed edge case scenarios