Control of Connected and Automated Vehicles: State of the Art and Future Challenges

Jacopo Guanetti; Yeojun Kim; Francesco Borrelli

Source-linked AI summary

Control of Connected and Automated Vehicles: State of the Art and Future Challenges

Jacopo Guanetti, Yeojun Kim, Francesco Borrelli

arXiv:1804.03757v1Apr 11, 2018eess.SY

TL;DR

CAV control and planning must address limited awareness of other road users while improving safety and energy efficiency. This paper proposes an architecture and surveys algorithms across its functional blocks, identifying interactions, opportunities, and challenges.

Problem
Limited awareness of other road users constrains autonomous driving, motivating connected control and planning for safer and more energy-efficient transportation.
Method
The paper defines a CAV control and planning architecture and surveys state-of-the-art approaches across controls, planning, eco-driving, coordination, and routing.
Results
The survey frames existing algorithms within the architecture and identifies their roles, mutual interactions, promising optimization-based approaches, and future challenges.
Takeaways & Limitations
CAV control and planning research requires clearer functional scopes and more investigation of interactions between blocks.
Takeaways & Limitations
Most surveyed technologies lack experimental validation, and representative real-world testing scenarios remain difficult to select.

Abstract

from arXiv · show

Autonomous driving technology pledges safety, convenience, and energy efficiency. Challenges include the unknown intentions of other road users: communication between vehicles and with the road infrastructure is a possible approach to enhance awareness and enable cooperation. Connected and automated vehicles (CAVs) have the potential to disrupt mobility, extending what is possible with driving automation and connectivity alone. Applications include real-time control and planning with increased awareness, routing with micro-scale traffic information, coordinated platooning using traffic signals information, eco-mobility on demand with guaranteed parking. This paper introduces a control and planning architecture for CAVs, and surveys the state of the art on each functional block therein; the main focus is on techniques to improve energy efficiency. We provide an overview of existing algorithms and their mutual interactions, we present promising optimization-based approaches to CAVs control and identify future challenges.

1. Introduction … Vehicle Powertrains

The paper frames CAVs as a means to improve transportation safety, efficiency, awareness, and cooperation by combining automated driving with connectivity. It surveys a control and planning architecture centered on energy-efficient real-time control, planning, coordination, and routing.

1. Introduction: Human error drives many road accidents, while congestion reduces transportation efficiency; reducing reliance on human drivers could improve travel-time use and safety.Autonomous driving is presented as a response to both safety and congestion problems.
1. Introduction: CAVs extend automation and connectivity by improving environmental awareness and enabling applications such as augmented awareness, platooning, cooperative maneuvers, and signalized-intersection assistance.Fast V2V communication is especially important for making full use of connectivity.
1. Introduction: The survey reviews a CAV control and planning architecture focused particularly on energy efficiency across real-time control, motion planning, eco-driving, multi-vehicle coordination, and routing.The scope emphasizes vehicle controls rather than traffic control.
2. System Components: Successful CAV deployment depends on both on-board instrumentation and the surrounding environment, including infrastructure and cooperative or non-cooperative road users.The surrounding environment includes signalized intersections, ramp meters, road signs, vehicles, cyclists, and pedestrians.
2.1. Connected and Automated Vehicle: A CAV combines automated driving with connectivity to vehicles, road users, infrastructure, and the cloud; powertrain control is important for energy-oriented applications but does not define CAVs.CAVs are distinguished by driving automation and connectivity.
Vehicles Automation: Automated driving controls longitudinal and lateral vehicle motion through interfaces with the powertrain and steering systems, using GPS, maps, and perception sensors.Perception sensors include lidar, radar, cameras, and ultrasonic sensors.
Vehicles Connectivity: Communication supports multi-vehicle cooperation, out-of-sight obstacle awareness, and forecasts through DSRC and cellular technologies, including 4G and 5G.DSRC applications include safety warnings, intersection assistance, traffic conditions, and toll payment; 5G also enables cloud services and potentially low-latency V2V/V2I communication.
Vehicle Powertrains: CAV motion control can improve energy usage, while powertrain systems can exploit forecasts from maps, perception, and communication.Auxiliary loads can significantly affect overall energy consumption, and non-safety-critical performance may be temporarily reduced to limit power consumption.

Highway infrastructure … Other Connected and Automated Vehicles

CAV infrastructure combines instrumented highways and intersections with connected urban, cloud, and cooperative-vehicle systems. These components provide sensing, traffic and map information, remote computation, and communication-enabled coordination, including platooning.

Highway infrastructure: Instrumented highways use in-roadway and over-roadway sensors to detect vehicles and support traffic monitoring, planning, and potentially active flow control.Technologies include loop detectors, magnetic detectors, magnetometers, cameras, radars, ultrasonic, infrared, and acoustic sensors.
Highway infrastructure: Highway sensors collect vehicle occupancy, speed, and type data for monitoring and planning road use.
Urban infrastructure: Instrumented intersections detect vehicle presence, estimate speed and turn movements, retrieve SPaT, and use these data for performance analysis, controller tuning, adaptation, and vehicle coordination.Controllers may use fixed cycles, traffic-responsive green times, or congestion-adaptive strategies; pedestrians may be included through cycle timing or button requests.
Urban infrastructure: Intersections can broadcast SPaT and geometry messages over DSRC using SAE J2735, but timing is uncertain when controllers do not use fixed cycles.
Urban infrastructure: Connected charging and parking infrastructure improves vehicle routing and pricing, while charging affects grid balancing and automated parking exploits urban surfaces more effectively.
Cloud infrastructure: Cloud-connected map services provide static road attributes and dynamic traffic, station, congestion, roadwork, and weather information, while historical data support routing and reference-velocity planning.Cloud services also provide remote computational power for dynamic routing and long-term trajectory optimization, partially reducing onboard computational requirements.
Other Connected and Automated Vehicles: Advanced multi-vehicle cooperation requires protocols beyond SAE J2735, and automated heavy-duty vehicles can form close platoons that reduce air drag resistance and fuel consumption.A standard for these more complex cooperation protocols has not yet been established.

Non-cooperative vehicles, cyclists and pedestrians … 4. On-board real-time control and planning

The paper frames CAV control as a layered architecture combining safety-critical, real-time on-board functions with longer-term remote planning and routing. Section 4 focuses on powertrain control, motion control, and motion planning, evaluated mainly through energy efficiency, comfort, throughput, and collision avoidance.

Non-cooperative vehicles, cyclists and pedestrians: Non-cooperative vehicles, cyclists, and pedestrians are handled primarily through onboard perception, while vehicle–smartphone communication can support safety cooperation with cyclists and pedestrians.CAVs do not substantially differ from other self-driving vehicles when road users are non-cooperative.
3. Connected and Automated Vehicle Control Architecture: Figure 2 presents a safe and energy-efficient CAV control architecture with both on-board and remote functional blocks.The on-board layer is safety-critical and real-time, whereas the remote layer performs longer-term, performance-oriented computations.
3.1. Real-time control and planning: The on-board real-time layer interfaces with actuators, collects sensor measurements, and executes control and planning computations needed for reliable responses to unpredicted events.Its algorithms include powertrain control, motion control, and motion planning.
Powertrain control: Powertrain control regulates engines, electric motors, and gear shifting to meet vehicle power demand while affecting tank-to-wheel energy conversion.Reactive methods select operating points from current demand, while forecasts can improve energy efficiency.
Motion control; Motion planning: Motion control executes higher-level longitudinal and lateral motion references in closed loop through interfaces with the powertrain and steering systems.It affects safety and wheel-to-distance energy conversion, while the real-time planning block covers maneuver, path, and trajectory planning with context-dependent boundaries.
Battery charge planning; Eco-driving and coordination: Long-term battery charge planning can prevent suboptimal stored-energy use by predicting electric-vehicle range and triggering route replanning or charging stops when needed.Eco-driving and coordination compute reference velocity trajectories using route information, forecasts, and constraints such as trip time, maximum velocity, and green phases; platooning can reduce gaps and aerodynamic drag.
3.3. What is not covered in this survey: Real-time planning and control depend on vehicle feedback, relative position and velocity, moving-obstacle predictions, and fused GPS, camera, radar, lidar, and DSRC data.The survey limits its scope to control, planning, routing, and coordination layers rather than perception, localization, and environment prediction.
3.4. How to read this survey; 4. On-board real-time control and planning: Section 4 reviews powertrain control, motion control, and motion planning using energy consumption, passenger comfort, road throughput, and collision avoidance as performance and safety criteria.The survey proceeds bottom-up, emphasizes optimization-based methods, and considers automation, cooperation, forecasts, connectivity, and coordination between on-board and remote layers.

Literature review · Challenges and opportunities for CAVs

The literature review covers gear-shifting, engine on/off, and energy-management control across fuel-powered, electric, hybrid, and plug-in hybrid vehicles. For CAVs, automation, sensing, communication, and improved forecasts create opportunities to coordinate these controls and reduce energy-wasteful events.

Literature review: The surveyed powertrain-control problems are gear shifting, engine on/off control, and energy management across fuel-powered, electric, hybrid, and plug-in hybrid vehicles.Table 1 maps these problems to different powertrain configurations.
Literature review: Powertrain control commonly minimizes weighted fuel and battery-power consumption, with dynamic programming solving known driving profiles and MPC approximating policies online.Real-time operation must address unknown future power demand.
Literature review: Gear-shifting control trades higher engine efficiency at low speed and high torque against drivability, while finite gear dynamics motivate discrete optimization and rounding-based approximations.Joint formulations combine gear shifting with energy management and, in some work, engine on/off control.
Literature review: Energy management allocates hybrid-vehicle power demand between the internal combustion engine and electric motor, subject to battery terminal-charge constraints.Plug-in hybrids add a fuel-versus-electricity trade-off because battery charge can be fully utilized.
Literature review: Real-time energy-management methods address unknown driving schedules through ECMS update laws, robust control, stochastic dynamic programming, and MPC using forecasts and constraints.Reference-state-of-charge generation and long-term route information are important for managing battery discharge.
Challenges and opportunities for CAVs: CAV automation, environmental awareness, and vehicle–infrastructure communication can make future velocity, wheel torque, and power-demand forecasts more reliable.These capabilities complement forecasts used by gear-shifting, engine on/off, and energy-management algorithms.
Challenges and opportunities for CAVs: In gear-shifting and engine on/off control, improved forecasts can help avoid energy-wasteful switching and shifting events, whose decisions depend on future conditions.Energy management benefits from both short-term and long-term forecasts of future demand.

Literature review

The literature review organizes longitudinal CAV control by the external information used, covering predictive, adaptive, urban, and cooperative adaptive cruise control before discussing lateral control. It highlights model predictive control, communication, and platoon design as central tools and challenges.

Longitudinal control: Longitudinal control approaches are organized by external information: predictive cruise control uses remotely computed reference velocity, adaptive cruise control uses perception, urban cruise control uses infrastructure communication, and CACC uses vehicle communication.The review then turns to lateral control.
Predictive cruise control: Predictive cruise control generates reference velocities from static or slowly changing preview information, often using cloud-aided optimization.Examples include road grade, speed limits, and traffic speed.
Adaptive cruise control: ACC adjusts speed to avoid collisions, while MPC can jointly support safety and performance through persistent-feasibility conditions.A control-invariant terminal set can ensure stability and persistent feasibility, although computing such a set is difficult with nonlinear dynamics and time-varying, non-convex constraints.
Urban cruise control: V2I signal-phase and timing information enables MPC-based urban cruise control strategies that have shown substantial energy savings compared with standard ACC.These strategies are particularly well addressed in arterial scenarios with successive traffic lights.
Cooperative adaptive cruise control: CACC uses V2V communication to exchange acceleration and forecasts, supporting distributed platoon control whose design depends on dynamics, information flow, local controllers, and formation geometry.Platooning also involves safety, string stability, throughput, aerodynamic drag, and a trade-off between reduced spacing and powertrain efficiency.

Challenges and opportunities for CAVs

CAV control can operate across highways, urban roads, and rural roads using finite-horizon optimal control, but cooperative driving still faces heterogeneous communications, delays, packet losses, and complex dynamics. Comprehensive frameworks for these systems and for safety–robustness trade-offs involving energy consumption and throughput remain lacking.

Open challenges: Cooperative driving controls must address diverse communication topologies and protocols, communication delays, packet losses, and complex vehicle dynamics.Although these systems have been systematically analyzed at least for highway platooning, a comprehensive framework is still lacking.
Open challenges: Safety and robustness requirements, including string stability, must be balanced against energy consumption and road throughput, but comprehensive analysis remains unavailable.The value of forecasts in this trade-off is not yet entirely clear; CACC commonly communicates the preceding vehicle’s velocity and acceleration, while V2V can share extended forecasts.
Longitudinal control: A CAV can implement longitudinal control across highways, urban roads, and rural roads through a finite-horizon optimal-control formulation in the time domain.The formulation uses state, input, and forecast variables to represent distance, vehicle speed, wheel force, braking torque, and the preceding vehicle’s velocity.
Longitudinal control: The stage cost trades control effort against velocity- and distance-tracking errors, while constraints limit actuators, speed, acceleration, and collision-avoidance distance.References are set by eco-driving when the road ahead is free and by the preceding vehicle otherwise; state constraints are dynamically shaped by measurements and forecasts.
Longitudinal control: A forecast-based terminal set can allow the ego vehicle to avoid preceding-vehicle collisions without braking throughout the forecast horizon.The property is defined for Nf >> N when the predicted terminal state lies in XN, and the approach was evaluated experimentally with a sinusoidal preceding-vehicle velocity profile.

Literature review

The literature review covers CAV decision making and motion planning, from behavioral choices to time-parameterized trajectories, while emphasizing uncertainty, computational tractability, and communication-enhanced awareness. It surveys established computational techniques and the potential of distributed perception among connected agents and infrastructure.

Decision making: Decision making covers vehicle behavior choices and translating those choices into CAV trajectories.Examples include lane keeping or changing, stopping at intersections, and yielding to pedestrians.
Decision making: Uncertain intentions of non-CAV road users make estimation, prediction, and learning central to behavioral decision making.Heuristic rules can implement vehicle behavior, but public-road uncertainty limits their sufficiency.
Motion planning: Path planning searches future vehicle motion in configuration space, whereas trajectory planning searches for a time-parameterized solution using current and predicted surrounding-object states.Both planning problems are commonly formulated as optimization problems.
Motion planning: Optimal path and trajectory planning are both PSPACE-hard in general, motivating tractable approximations and scenario-specific methods.Path-planning methods include variational, graph-search, and incremental-search techniques; trajectory planning can use time-domain variational methods or add a time dimension to path planning.
Connected perception: Communication can improve environmental awareness by sharing agent states, predicted motion, and detected obstacles across CAVs, roadside units, cyclists, and pedestrians.The review describes this distributed perception potential as capable of significantly outperforming advanced perception based only on on-board sensors.

Challenges and opportunities for CAVs

CAV planning must be tailored to vehicle and environmental constraints while balancing motion-planning complexity against longitudinal and lateral control. Key opportunities and challenges arise from extended V2V/V2I sensing, coordinated multi-agent motion, and non-convex collision avoidance.

Motion planning depends on the vehicle and environmental constraints, with a trade-off between planning complexity and longitudinal and lateral control complexity.The planning approach must be selected for the specific application and anticipated conditions.
Perception, environmental prediction, and models of other agents strongly influence decision making and motion planning, while V2V and V2I can extend sensing capabilities.Communication can provide advantages in scenarios where perception systems are insufficient.
Multi-agent CAV planning must support safe, coordinated formations by addressing platoon merging, dismantling, lane changing, and collision avoidance.These problems must be solved feasibly for the entire formation rather than for a single vehicle alone.
An optimization-based trajectory-planning formulation incorporates CAV dynamics and actuator limits for collision avoidance while minimizing weighted time and input effort.The example considers one CAV moving in two-dimensional space among multiple obstacles using a kinematic bicycle model.
Collision-avoidance constraints are generally non-convex and non-differentiable, but they can be reformulated as smooth, non-conservative nonlinear constraints.The reformulation applies when the CAV and obstacles are representable as finite unions of convex sets.

5. Remote planning and routing · Literature review

Remote planning and routing uses route, traffic, and forecast data for long-term computations that improve overall trip performance, while coordinating with on-board functions. The literature review covers battery charge planning, eco-driving, and eco-routing, emphasizing adaptive charge references and repeated computation when needed.

5. Remote planning and routing: Remote planning and routing exploits route and traffic data to maximize CAV overall trip performance across energy consumption, trip time, driver convenience, and road throughput.Coordination with on-board functional blocks is fundamental to achieving the desired performance improvement.
5. Remote planning and routing: The literature review organizes on-board methods into battery charge planning, eco-driving, and eco-routing, while noting that practical approaches often cross these boundaries.Eco-driving approaches are discussed separately for isolated vehicles and groups of vehicles.
5. Remote planning and routing: Many reviewed algorithms can run in the cloud, on a CAV, or at a roadside coordinator, depending on the application.Reference generation and routing may occur at departure, but recomputation during the trip is often advised or required.
Literature review: Driving range and charging time remain pressing electrified-powertrain problems despite battery advances, while route, traffic, and weather forecasts improve electric-range estimates.These data sources can substantially improve the accuracy of estimated electric driving range.
Literature review: Electric-vehicle battery depletion can be predicted more accurately and potentially counteracted by limiting auxiliary or traction power or planning charging-station stops.The cited strategies address depletion through power limitation or route-based charging planning.
Literature review: HEV charge should remain bounded throughout the trip and reach a target value at the end, but constant reference-charge signals can struggle with charge constraints during large altitude variations.Most real-time energy-management approaches postulate a reference charge signal, usually chosen constant.
Literature review: For plug-in HEVs, convex-program and dynamic-programming charge references using logged route velocity and altitude achieved similar performance and clearly outperformed CDCS.The routes were assumed to be commuting routes, and the final charge must be at least a minimum level to avoid deep discharge.
Literature review: HEV reference-charge trajectories can use elevation profiles and speed limits or average traffic speed to maintain prescribed limits and maximize recuperation during deceleration and downhill segments.The goals are intertwined, particularly for HEVs with small batteries, because charge must be dynamically controlled to fully exploit recuperation.

Challenges and opportunities for CAVs · Literature review · Challenges and opportunities for CAVs

The review presents integrated CAV control as an opportunity to improve energy efficiency through coordinated eco-driving, battery-charge planning, and vehicle–infrastructure interaction. Key challenges include uncertain traffic and signal behavior, while robust and forecast-based optimization offers promising directions.

Challenges and opportunities for CAVs: Integrated CAV architectures can use accurate future power-demand forecasts from eco-driving to optimize battery-charge depletion.This extends beyond approaches relying only on static route information such as road grade and speed limits.
Challenges and opportunities for CAVs: Electric and plug-in hybrid planning can incorporate charging-station stops, charging and waiting times, dynamic pricing, and smart-grid interactions.Related opportunities include grid balancing and interaction with smart grids.
Example: battery charge planning for a connected plug-in hybrid electric vehicle: Battery-charge planning is formulated as a finite-horizon optimal-control problem using battery energy, motor and engine torques, and forecasts of speed and auxiliary power.The eco-driving block can provide the reference-speed forecast, while auxiliary-power forecasts may use weather data and an onboard model.
Example: battery charge planning for a connected plug-in hybrid electric vehicle: Optimal battery management blends motor and engine usage, contrasting with the charge-depleting/charge-sustaining pattern observed in measured commute data.The comparison uses a typical Bay Area commute with driving data measured on a plug-in hybrid test vehicle.
5.2. Eco-driving for isolated CAVs: Eco-driving computes minimum-energy trajectories using route information, long-term forecasts, trip-time and speed constraints, stops, and intersections for isolated CAVs.Reference velocities can combine road topology, grade, curvature, speed limits, traffic speed, and weather forecasts.
Literature review: 5–15 % fuel economy improvement was observed when cloud-based dynamic programming optimized reference velocity and a human driver tracked it.The optimization considered road geometry, grade, traffic information, and an accurate vehicle and powertrain model.
Literature review: Signalized-intersection eco-driving faces uncertainty from traffic, queues, pedestrians, and adaptive signal phases, which few cited works explicitly model.A probabilistic signal-timing forecast and a chance-constraint formulation address imperfect phase-duration forecasts.
Example: eco-driving using signal timing data: Robust signal-aware solutions provide adjustable conservatism through η_i, whereas deterministic planning may cross an intersection near a red-to-green phase switch.The level of conservatism is tuned by changing η_i in the robust formulation.

Literature review

The literature review surveys multi-vehicle coordination problems for CAVs on autonomous roadways and in platoons. It covers speed harmonization, merging and intersections, platoon coordination, formation, maneuvers, and route-based clustering.

Scope: Research on multi-vehicle trajectory planning for CAVs addresses autonomous roadways, including speed harmonization, merging roadways, autonomous intersections, and platoon coordination.The review positions these as selected coordination problems within a broader literature spanning autonomous robots, unmanned aerial vehicles, and marine vehicles.
Coordination on autonomous roadways: Speed harmonization controls individual CAV speed trajectories before speed-reduction zones, with analytical solutions and comparisons against human driving, variable speed limits, and patrol vehicles.Without lane changes, the safety requirement is avoiding rear-end collisions; with perfect trajectory tracking, the problem can be solved in a fully decentralized manner.
Coordination on autonomous roadways: Merging roadways and autonomous intersections require smooth coordination that avoids stop-and-go driving, rear-end collisions, and lateral collisions where vehicle streams merge or cross.Recent research studies both problems under the assumption that all vehicles on the road are CAVs, and the literature includes multiple optimization-based formulations.
Platoon coordination: Platoon coordination is mainly distributed for longitudinal control, but centralized coordination remains necessary for formation geometry and information exchange.Platoon agents are assumed to be CAVs even when all road vehicles are not; formation geometry includes spacing policy, cruising speed, and vehicle ordering.
Platoon coordination: Platoon research covers formation, dismantling, merging, splitting, lane changing, and route-based clustering using hierarchical control and optimal-control approaches.One approach constrains a trail platoon’s velocity below a maximum safe velocity determined by spacing and lead-platoon velocity; other work clusters vehicles by routes and departure and arrival times.

Challenges and opportunities for CAVs

Deploying CAV coordination on public roads requires real-time planning that reacts to surrounding traffic, accommodates interactions with non-CAVs, and addresses the effects of partial CAV penetration. Key challenges include balancing platoon safety and robustness against energy consumption and road throughput while making complex coordination problems computationally tractable.

Public-road deployment: Public-road deployment requires real-time planning that reacts to surrounding traffic and supports CAV interaction with other vehicles.Open questions include reformulating coordination problems and assessing performance under partial CAV penetration.
Platoon coordination: Platoon coordination lacks a unified framework for different communication topologies and must balance safety and robustness, including string stability, with energy consumption and road throughput.
Platoon coordination: The coordination formulation allows vehicle-order changes and flexible platoon formation while enforcing collision avoidance, actuator limits, speed limits, and switching constraints.The objective is to minimize total wheel energy over finite-horizon longitudinal trajectories.
Computational tractability: Platoon trajectory optimization is complex because it combines mixed-integer nonlinear dynamics with a high-dimensional state space.A distributed receding-horizon approach using smooth dynamics has been proposed to approximate the optimal solution computationally.

Literature review

Energy-optimal routing is a shortest-path problem under deterministic, time-invariant consumption models, but model choice, travel-time tradeoffs, electrification, charging, and signalized networks add important complexity. Research therefore spans model-based, data-driven, constrained, dynamic-programming, and optimization-based eco-routing methods.

Energy-optimal routing models: Under deterministic, time-invariant energy models, energy-optimal routing reduces to a shortest-path problem.Studies use comprehensive emissions, data-driven, and vehicle longitudinal-dynamics models, alongside time-varying and uncertain models.
Energy-optimal routing models: Eco-routing methods compute edge costs from data or vehicle models and are compared with shortest and fastest routes in realistic SUMO-generated traffic.The energy-consumption model is identified as critical for the comparison.
Routing tradeoffs: Because eco-routes can be time consuming or lengthy, multi-objective and constrained shortest-path algorithms balance fuel, travel time, and distance or impose travel limits.Constrained formulations minimize fuel consumption subject to maximum travel-time and/or travel-distance constraints.
Electrified-powertrain routing: Electrified-powertrain routing must track battery energy for driving-range guarantees, while regenerative braking can create negative energy costs on some road segments.These features require added complexity and modifications to standard routing algorithms.
Electrified-powertrain routing: Minimum-time routing with limited energy and charging stops has been solved using mixed integer non-linear programming and dynamic programming, including a multi-vehicle congestion extension.An alternative flow-optimization formulation is proposed to mitigate computational complexity in the multi-vehicle case.
Signalized-network routing: Signalized-network eco-routing models velocity trajectories as uncertain and estimates edge costs with a microscopic vehicle-emission model within a Markov decision process.Velocity-trajectory estimation over links is more challenging than when using historical data alone.

Challenges and opportunities for CAVs · 6. Conclusion and outlook

The paper frames CAV control and planning as a system-level architecture whose functional-block interactions and experimental validation remain incomplete. It also highlights eco-routing opportunities, including energy-aware routing and more systematic treatment of uncertainty.

Example: eco-routing for a plug-in hybrid electric vehicle: Eco-routing formulations represent routes as paths through a graph of intersections and road segments between an origin and destination.Nodes represent important road locations, while edges represent connecting road segments.
Example: eco-routing for a plug-in hybrid electric vehicle: The minimum-energy route minimizes an edge-cost sum combining fuel energy E_f and battery energy E_q.The edge cost depends on the current and next nodes and the current battery energy.
Example: eco-routing for a plug-in hybrid electric vehicle: Vehicle speed, road grade and curvature, powertrain dynamics, and onboard control strategies make fuel and battery energy consumption complex functions.For plug-in hybrid vehicles, energy use also depends strongly on the current battery state x_k.
Example: eco-routing for a plug-in hybrid electric vehicle: Battery constraints can be enforced through a safe operating range and a target destination charge, with charging-station stops accommodated along the route.The terminal charge affects charging time until the next trip.
Example: eco-routing for a plug-in hybrid electric vehicle: 10.03 km and 3.22 kWh characterize the minimum-energy route, versus 9.72 km and 4.23 kWh for the minimum-distance route.Both routes are compared for an origin and destination in the Berkeley area using the same simple plug-in hybrid vehicle model.
6. Conclusion and outlook: The proposed system-level architecture organizes existing CAV control and planning algorithms across hierarchical functional blocks.The survey covers algorithms and technologies for connected and automated vehicles and identifies a possible control and planning architecture.
6. Conclusion and outlook: Interactions and possible integration between functional blocks have not been exhaustively investigated.The paper emphasizes that algorithms need well-defined scopes while their cross-block relationships remain insufficiently studied.
6. Conclusion and outlook: Experimental validation is lacking for most technologies, although advanced algorithms can increasingly be deployed on real vehicles.Public-road testing remains challenging, and representative real-world testing scenarios are a non-trivial open question.