Automated vehicles (AVs) provide great controllability that can lead to lower fuel use and traffic delays in transportation networks. Prior work by the PI has demonstrated that properly optimized AV platoons can reduce both travel time and emissions by as much as 40% when traveling through an isolated intersection. But the method used to achieve these results—i.e., optimal trajectory control—is complex and time consuming to solve, hence not amenable for real-time operations. Such complex problems, on the other hand, may be perfect applications for AI (machine learning). In this research, we plan to 1) usie machine learning to optimize the AV platoon trajectories to reduce fuel use and travel delay, and 2) consider a mixture of personal cars and buses in the AV platoon, so that the system can be optimized for person-centric rather than vehicle-centric mobility and footprint. We will first identify, from the literature, proper energy consumption models for both passenger cars (gasoline) and buses (natural gas in our case, to align with local practices; electric vehicles are not considered in the current work). Then we will develop proper reward functions for the reinforced machine learning algorithms and design the learning cases. The training of the machine learning algorithms will be implemented through microscopic simulation (VENTOS/SUMO). We'll then evaluate the machine learning optimization through several test scenarios under different demand levels of buses and passenger cars.