Sustainable Computing for Traffic Automation

Computing-enabled Automated Traffic Systems (ATS) are rapidly gaining popularity for autonomous vehicles and smart infrastructure, leading to increased power consumption and energy costs. This paper first demonstrates that the computational demands of ATS substantially elevate energy consumption. Fully autonomous vehicles alone require an additional 205,000 kWh of electricity over a typical 12-year lifespan, and associated smart traffic infrastructure consumes approximately three times more energy. The primary driver of this increased energy usage is the intensive sensing and computing functions powered by deep neural networks, which account for about 98% of total utility-phase energy costs and 95% across the entire lifespan. By integrating state-level energy portfolios and policies, we develop and make available a new accounting framework and dynamic model to predict the carbon emissions trajectory from 2025 to 2100 for California, Ohio, and the United States average. We find that proactive emission mitigation strategies can be categorized and then addressed by short-term and long-term policy actions. Hardware computing efficiency improvements yield the strongest short-term emission reductions, whereas long-term reductions can be achieved through the rapid adoption of electric autonomous vehicles and the early retirement of older, inefficient vehicle fleets.