A Collaborative Decision-Making Framework for Multilane Ramp-Merging in Heterogeneous Traffic Using Improved Multi-Agent Deep Deterministic Policy Gradient

Ramp-merging areas on freeways are critical nodes for traffic flow transitions but often suffer from low efficiency and elevated safety risks because of multi-vehicle conflicts. This study proposes a distributed decision-making architecture for multilane ramp lane-changing and merging scenarios and develops a Maneuver-Enhanced Multi-Agent Deep Deterministic Policy Gradient (ME-MADDPG), which unifies lane-changing and merging maneuvers in a multi-agent reinforcement learning framework. A relatively stationary interaction environment and localized actor–critic structure adapt the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) to continuously changing vehicle sets, while region-specific state, action, and reward designs distinguish flow redistribution in the upstream Scheduling zone from safe gap acceptance in the downstream Merging zone. A multidimensional state space representing connected and automated vehicles (CAVs), human-driven vehicles (HDVs), and traffic conditions is coupled with a dynamic-weight multiobjective reward integrating safety distance constraints, merging delay, acceleration smoothness, and cooperative behavior. Microscopic simulations of multilane mixed CAV–HDV traffic demonstrate the effectiveness of the strategy. Compared with first-in, first-out (FIFO) merging and single-agent deep deterministic policy gradient, ME-MADDPG reduces total delay, improves safety distance compliance, and increases successful cooperative merges over a wide range of traffic densities. Relative to the original MADDPG, the method further improves lane utilization balance, increases mean time-to-collision, mitigates longitudinal and lateral conflicts, and achieves shorter single-step decision time with slower degradation as density increases. These results indicate that ME-MADDPG can support real time cooperative control and enhance both efficiency and safety in dense multilane ramp-merging environments.