Analyzing Emissions and Energy Efficiency at Unsignalized Real-world Intersections Under Mixed Traffic Control

TL;DR

This paper evaluates emissions reductions from mixed traffic control at unsignalized intersections using RVs guided by reinforcement learning. Modeling the problem as a multi-agent POMDP solved with Rainbow DQN, the authors train RVs in SUMO across four real-world Colorado intersections with real demand, measuring fuel, $CO_2$, $CO$, $HC$, and NOx emissions under 10–100% RV penetration and comparing to HVs at signalized intersections. Results show consistent reductions in fuel, $CO_2$, and NOx at all penetration levels (up to 27–28%), and up to 42–43% reductions in $CO$ and $HC$ at higher penetrations, with network-wide ripple benefits even when control is exercised only at intersections. The work highlights that eco-driving-like benefits emerge naturally from mixed traffic coordination, suggesting unsignalized intersections as a promising avenue for sustainable traffic control and informing future enhancements such as continuous actions and cross-intersection communication.

Abstract

Greenhouse gas emissions have dramatically risen since the early 1900s with U.S. transportation generating 28% of U.S. emissions. As such, there is interest in reducing transportation-related emissions. Specifically, sustainability research has sprouted around signalized intersections as intersections allow different streams of traffic to cross and change directions. Recent research has developed mixed traffic control eco-driving strategies at signalized intersections to decrease emissions. However, the inherent structure of a signalized intersection generates increased emissions by creating frequent acceleration/deceleration events, excessive idling from traffic congestion, and stop-and-go waves. Thus, we believe unsignalized intersections hold potential for further sustainability improvements. In this work, we provide an emissions analysis on unsignalized intersections with complex, real-world topologies and traffic demands where mixed traffic control strategies are employed by robot vehicles (RVs) to reduce wait times and congestion. We find with at least 10% RV penetration rate, RVs generate less fuel consumption, CO2 emissions, and NOx emissions than signalized intersections by up to 27%, 27% and 28%, respectively. With at least 30% RVs, CO and HC emissions are reduced by up to 42% and 43%, respectively. Additionally, RVs can reduce network-wide emissions despite only employing their strategies at intersections.

Figures (4)

• Figure 1: LEFT: One of the complex real-world unsignalized intersections Wang et al. wang2023learning train on with mixed traffic (RV in red and HV in white). RIGHT: Wang et al. show RVs significantly reduce wait times compared to HVs with signalized intersections past 60% RV penetration rate. Wait time results are averaged across four unsignalized intersections.
• Figure 2: We perform our emissions analysis on four, four-way intersections originating from Colorado Springs, CO, USA. Each of the intersections has been accurately recreated in SUMO behrisch2011sumo. We label the intersections as 229, 499, 332, and 334, from north to south geographically. The traffic flows for each intersection come from real-world turning count data, meaning they represent real-world traffic demand on complex intersections wang2023learning. The RVs (red; human-driven vehicles are white) learn to control and coordinate mixed traffic in the intersection to reduce wait times, while preventing any conflicts. The intersections are learned on individually; however, as the intersections simultaneously exist during learning, what the RVs do at one intersection can affect the rest. We investigate if the learned mixed traffic control strategies also reduce emissions at the intersections.
• Figure 3: Results for fuel consumption and CO$_2$, CO, HC, and NOx emissions averaged across the four unsignalized intersections. We measure in 10% increments from 10% to 100% RV penetration rate. The dotted, black line represents the baseline HVs with signalized intersections performance. For fuel consumption, CO$_2$ emissions, and NOx emissions, every RV penetration rate outperforms the baseline by up to 27%, 27%, and 28% (at 70% RV penetration rate), respectively. CO and HC emissions requires at least 30% RV penetration rate to outperform the baseline. The largest improvement for CO and HC emissions is 43% and 42% (both at 100% RV penetration rate), respectively. We observe that despite not outperforming HVs with signalized intersections on wait time until 50% (see Fig. \ref{['fig:efficiency']}), we are still able to reduce generated emissions at lower RV penetration rates. Overall, the RVs employ mixed traffic control strategies that naturally reduce generated emissions despite not being trained to intentionally reduce emissions.
• Figure 4: Acceleration profile comparisons between HVs with traffic signals and either 20%, 40%, 60%, or 100% RV penetration rates at Intersection 229 and Intersection 334. Intersection 229 is selected due to it having a different topology from the other intersections, and Intersection 334 is chosen as a representative for Intersections 499 and 332 due to all three having similar topologies. Each acceleration presented is the average across all vehicles present at the intersections for each second. Values have been smoothed using a moving average with window size of $10$. At 20%, 40%, 60% RV penetration rates for both intersections, we observe more stable acceleration rates around 0 compared to HVs with traffic signals that see periods of sudden, high accelerations. These more stable accelerations results in, on average, less fuel consumption being used. For Intersection 229 at 100% RV penetration rate, we observe spikes in accelerations similar to the baseline; however, we believe generated emissions are still less due to less excessive idling at the intersections. For Intersection 334, 100% RV penetration rate sees more accelerations around 0 compared to at Intersection 229, leading to less generated emissions. Overall, the RVs at each considered RV penetration rate exhibits behavior that can lead to reduced emissions.