Coordinating variable speed limits and managed lanes at on-ramp bottlenecks: Policy design, performance, and implementation implications

Recurrent on-ramp bottlenecks remain a major source of delay and unreliability on urban freeways. We evaluate a coordinated variable speed limit (VSL) and managed lane (ML) policy that (i) meters mainline vehicle speeds by section and (ii) dynamically selects ML termini upstream and downstream so that ML remains outside the merge influence area to preserve weaving capacity. The research is established for fully connected and automated vehicles, where vehicles are expected to adapt to VSL and ML consistently. The model is solved with an option-critic controller of deep reinforcement learning, which selects VSL values for each mainline cell and on-ramp cell as well as ML termini at each control step, using measured road occupancy and vehicle speed states, where a comprehensive reward is developed to balance person delay and ramp queue dissipation. In microsimulation, coordinated VSL and ML reduces average vehicle delay, passenger delay, stop time, and CO2 by up to 57 %, 48 %, 42 %, and 37 %, respectively, compared with no-control baselines; gains are most robust at moderate flows between 4500 and 5500 veh/h and priority shares 20–60 %, while passenger delay improves most at 40–50 % priority share. Lane-placement experiments show that ML performs best on the outside (ramp-adjacent) lane with its termini offset from the merge area. Findings may translate into design guidance: keep ML openings out of the merge influence area and coordinate VSL value to tune inflow into the weaving zone.