13
learning by minimizing predicted domain labels, while others b) Individual vehicle: On top of powertrain control, a
aim to minimize the discrepancy between learned features vehiclecanbebettercontrolledbyleveragingCVinformation.
from the source and target domains [117]. First, vehicle speed can be optimized based on predicted
2) Connected and Autonomous Vehicles: Emerging smart future traffic conditions. For example, a CAV can receive
transportation technologies such as connected and automated future SPaT to arrive at an intersection during green light
vehicles (CAVs) provide new opportunities to improve safety, and avoid waste due to braking. This is referred to as Eco-
energy efficiency, sustainability, and mobility of the automo- approach applications [319], [320], [321]. Second, vehicle
tive and transportation sector [297], [298], [299]. Connected routingcanbeplannedtoreducetraveltimebasedonreal-time
vehicles (CVs) are essentially edge IoT devices that can con- information of the traffic flow speed [322], [323].
nect to the internet or other edge devices such as surrounding c) Multi-vehicle cooperative driving automation (CDA):
CVs, smart infrastructures (e.g., traffic signal lights), and An individual vehicle-based vehicle control strategy can be
other connected road users (e.g., pedestrians, cyclists). CVs selfish and compromise the performance of other vehicles.
are equipped with onboard communication devices that are CDAstrategiescanbedevelopedtooptimizemultiplevehicles
enabled by communication technologies [300], [301] such as to achieve global optimality for both the overall traffic and
5G,C-V2X,ordedicatedshort-rangecommunication(DSRC). eachvehicle(agent).Applicationssuchascooperativeadaptive
Thus, real-time communication can be established to obtain cruise control [324], [325], [326], [327], cooperative merging
information on other vehicles’ position and speed, signal [328], [329], [330], [331], speed harmonization [332], [333]
phase and timing (SPaT), routing and road curvatures, etc. have been proposed in the literature. In addition, when con-
In addition to connectivity, a CV equipped with vehicle au- sideringmultipleCAVs,cooperativeperception[334]becomes
tomationtechnologiesbecomesaCAVwhosereal-timevehicle possiblewhereCAVscollaborativelyperceivetheenvironment
states and motions can be precisely controlled. SAE has to extend perception beyond local sensing capability, improve
defined different levels of automation [302], from advanced safety, and reduce the computational power needed.
driver assistance systems (ADAS) such as adaptive cruise d) Traffic infrastructure: In addition to vehicle-centered
control (ACC), lane keeping, to fully automated vehicles. control strategies, traffic infrastructures can be further opti-
CAV technologies not only leverage real-time communication mized as CVs provide real-time traffic information regarding
to significantly extend the line-of-sight of a human driver arrival time, demand, routing, etc. The literature has studied
or traditional onboard sensors to receive traffic information various applications, including smart signal control strategies
hundreds of meters away [303], but also bring in new control [335], [336], [337], variable speed limit [338], [339], ramp
means through vehicle automation. Therefore, future traffic metering [340], [341], etc. Also, because of IoT, CVs or con-
conditions can be predicted more precisely to provide oppor- nected infrastructures can be used as remote sensing devices
tunitiesforproactivevehiclesortrafficcontrolstrategies[297]. to provide real-time traffic data. Proactive traffic management
ManyCAVtechnologieshavebeenstudiedintheliterature. strategies [342], [343] can be designed to improve overall
It is necessary to give an introduction to the state-of-the-art traffic performance, such as lane management, incident and
CAV applications before discussing how AGI can potentially emergency response, and provision of information and guid-
address challenges in the current literature. CAV technologies ance.
can be categorized into the following levels based on control It is anticipated that combining the above categories can
means: introduce further benefits. For example, both individual and
a) Powertrain: A vehicle’s powertrain control includes multi-vehicle control strategies can be integrated with power-
energy management strategy (EMS) and transmission gear train level control to improve performance for CAVs [307],
shifting control. EMS determines the battery-engine power- [319].Co-optimizedvehiclesandtrafficinfrastructurecontrol,
split of hybrid electric vehicles (HEVs), power-split of hy- such as integrated vehicle speed and signal light control,
draulic hybrid vehicles (HHVs), and multi-motor power-split can avoid conflicts between the two systems and maximize
of electric vehicles (EVs). Transmission gear shifting control energy efficiency and mobility [344], [345]. Recently, the
changes the torque and speed ratios between the engine and electrificationofvehiclesandtransportationhasattractedmore
thedriveline[304].Powertraincontrolstrategiesinproduction attention,andseveralnewopportunitieshavebeenproposedin
vehiclesareoftentunedconservativelyforthemostdemanding the literature. EVs have different powertrain architectures and
conditions to balance between efficiency and drivability. They can be controlled to improve powertrain efficiency. Electrified
are mostly rule-based and cannot adapt to varying driving vehicles generally have a faster response time to effectively
conditions [305], [306]. Many smart powertrain control ap- control and actuate the vehicles in highly dynamic driving
proaches have been studied in literature to leverage better conditions. EVs also have more electrical power onboard
predicted future conditions from CVs. Therefore, powertrain to support increased computational capabilities for complex
efficiency can be improved by proactively controlling gear driving tasks.
shiftingofconventionalvehiclesorpower-splitofHEVs[307], AGI can be a promising solution for the above CAV and
[308],[309],[310],[311],HHVs[312],[313],[314],andEVs smarttransportationapplicationsintheliterature.Alltheabove
[315], [316], [317]. For example, EMS can use more battery applications require an effective prediction of future traffic
power when it anticipates a HEV will decelerate during an conditions, which relies on the modeling and understanding
intersection as the battery power will be replenished from of the overall system. The system can include several vehicle
regenerative braking [318]. types with clearly different driving behaviors: human-driven