From Longitudinal Control to Networked CPS: Types, Methods, Field Layers, and Discourse Rotation in Vehicle Platooning

This post grows out of my ongoing Chinese review project on vehicle platooning. It is not the review itself. Instead, it asks a more basic question first: what do we miss if we still describe vehicle platooning as nothing more than a longitudinal car-following control problem?

My answer is: we miss quite a lot. Representative papers from 2025–2026 already show that, while safety, string stability, and cooperative control remain central, the field is now clearly expanding toward V2X, cloud control, cooperative perception, digital-twin testbeds, and a broader cyber-physical systems (CPS) framing 1 2 12 13.

So the most important change is not simply that “there are more controllers now.” The deeper change is that the object of research, its ontology, its methods, and even its problem-setting are shifting together. And to explain that shift adequately, it is not enough to place the four discourses side by side as static boxes; we also need to see how their rotation keeps pushing the field beyond its old longitudinal-control boundary. All citation numbers in the text are clickable and jump to the reference index at the end.

Why vehicle platooning needs to be reframed now#

When recent papers are read side by side, one pattern becomes difficult to ignore: vehicle platooning is no longer treated only as a problem of maintaining stable gaps in a line of vehicles. It increasingly appears as a systems problem shaped by communication, computation, sensing, infrastructure, and deployment conditions.

Recent surveys no longer discuss control in isolation. They explicitly place modeling, communication, control, and future directions in one frame 1. V2X safety research goes further and rewrites “safety” itself as a problem of collective intelligence across vehicles and infrastructure rather than ego-centric perception alone 2. Work on heterogeneous vehicle groups in ramp cloud-control systems, and on human-in-the-loop mixed digital twin platforms, shows the same outward movement: platooning is being absorbed into a broader systems vocabulary 12 13.

That is why a narrow review outline now feels insufficient. My current shorthand for the field is this:

Vehicle platooning research is moving from a longitudinal control problem toward a networked CPS problem organized by communication, computation, and control in coordination.

What kinds of research now make up the field#

Looking across the core and secondary papers from 2025–2026, at least five research types become visible. They are not isolated islands. Together they form the current intellectual map of the field.

Research type	Main concern	Representative papers
Survey and problem-framing	Define the object, scope, and future directions of platooning research	1, 2, 16
Stability and controller design	Guarantee safety and string stability under delays, switching, and uncertainty	3, 4, 10, 11
Robust predictive and data-driven control	Handle nonlinearity, mismatch, disturbances, attacks, and heterogeneity	5, 6, 12
Learning-enhanced and knowledge-informed methods	Bring DRL, self-learning, and knowledge distillation under engineering constraints	7, 8, 9, 15
System coordination and experimental platforms	Integrate vehicle, road, cloud, sensing, and testing infrastructures	2, 13, 14

1) Survey and problem-framing#

These papers matter because they decide what counts as the proper object of vehicle platooning research in the first place. The survey by Li et al. is important not just because it summarizes papers, but because it forces modeling, communication, control, and future prospects into one shared narrative 1. The Chinese monograph on communication-delay-aware truck platoon control performs a similar framing function from a more control-centered angle, organizing delay, topology, stability, and compensation into a systematic knowledge base 16.

In other words, these works do not merely describe the field. They teach readers how to see it.

2) Stability and controller design#

This remains the hardest and most classical line of work in platooning. Its core question is simple and demanding: how can safety and string stability be preserved under communication delay, uncertainty, switching, and changing formations 3 4 10 11?

ruan2025stability uses Lyapunov–Krasovskii tools and H∞ control to derive stability conditions under rate-free time-varying delay and uncertainty 3. tian2026mpcbased studies controller switching when communication delay exceeds a theoretical boundary, which is already a sign that the field is asking not only how to optimize control, but how to degrade gracefully at the system boundary 4.

3) Robust predictive and data-driven control#

If the previous line is centered on provability, this line is centered on the fact that the real world does not respect ideal models. Mixed traffic, nonlinear dynamics, disturbances, attacks, and mismatch all push platooning toward data-driven and robust predictive designs 5 6 12.

li2025robust combines T-S fuzzy modeling with robust MPC to deal with uncertain dynamics in mixed platoons 5. li2026robust goes further by combining Koopman lifting and reachability analysis to build a robust nonlinear data-driven predictive control framework under noise, disturbances, and attacks 6. shi2026robust places the problem inside ramp cloud control with heterogeneous vehicle groups, showing that “platoon control” is already stretching into a more general problem of coordinated trajectory tracking for heterogeneous groups 12.

4) Learning-enhanced and knowledge-informed methods#

The most interesting recent change is not simply that “deep learning has arrived.” The more important shift is that learning is now being asked to justify itself under stability, safety, and deployment constraints. Learning is not replacing control. It is negotiating for a constrained position within it.

pan2026reinforcement uses decentralized reinforcement learning for eco-safe control in mixed platoons 7. wang2025knowledgeguided introduces knowledge-guided self-learning under communication delays, with traffic stability, comfort, and energy cost still acting as hard engineering anchors 8. wang2025knowledgeinformed uses LLM-extracted traffic knowledge and distillation to build lightweight, stability-aware car-following models 9. Meanwhile, liu2025robust studies intention prediction under vehicle–infrastructure cooperative perception errors, showing that learning has also spread into perception and behavior modeling 15.

5) System coordination and experimental platforms#

This is the clearest sign that the field boundary is expanding. These works no longer stop at “the algorithm works in simulation.” They rewrite the experimental object itself: what kind of platform is needed, how cloud control is organized, how vehicles and infrastructure become one system, and how risk is conceptualized 2 13 14.

zhang2025cooperative organizes cooperative safety through an SPD framework, explicitly tying sensing, predictive reasoning, and coordinated decision-making together 2. dong2026multisource combines mixed digital twins and human drivers in the loop, folding physical vehicles, virtual vehicles, and human operators into one experimental circuit 13. gao2026concept shifts attention from control performance alone toward a broader language of system risk through driving risk entropy 14.

The method spectrum of the field#

Methodologically, vehicle platooning is not well described by a simple story of “traditional control versus AI.” What we actually see is a layered and entangled spectrum.

Control theory still sets the norm#

Whether a paper uses MPC, H∞ control, switching control, ADRC, Youla parameterization, or a data-driven predictive variant, it usually returns to the same unavoidable criteria: safety, string stability, robustness, and deployability 3 4 9 11. Control theory still provides the normative center of the field.

Pure model-based work is being rewritten by hybrid methods#

At the same time, purely idealized model-based approaches are being reworked into hybrid ones. Models, data, knowledge, and experimental environments are no longer kept separate. They increasingly appear as combinations of control backbones, data augmentation, learning-based compensation, and system-level validation 6 8 12 13.

Learning is entering under constraint#

This point matters. The more valuable learning papers are not those that merely claim “end-to-end works better,” but those that answer a harder question: how far can learning be disciplined by safety, stability, and deployment demands? That is what makes wang2025knowledgeinformed especially important: it inserts knowledge distillation and stability goals directly into the training process 9.

Validation has become a method problem#

Validation used to look like an auxiliary stage. Newer papers show that it is now part of the method itself. Bench tests, scaled experiments, real-world datasets, human-in-the-loop settings, and mixed digital twin platforms are no longer just “more evidence.” They are redefining what counts as adequate evidence for deployment claims 12 13 15 16.

Field, ontology, epistemology, and teleology: how the field is rewriting itself#

If research types and method families answer “what is being done” and “how it is being done,” this four-layer stack answers a different question: what world the field believes it inhabits, what it takes to be its real object, how that object can be known, and what ends organize the whole enterprise. At that level, vehicle platooning can no longer be treated as a narrow subtopic internal to control theory. It increasingly appears as a composite field at the intersection of control, traffic engineering, vehicular networking, V2X, cloud coordination, digital-twin experimentation, and CPS language 1 2 12 13 14.

Layer	Current shift	Typical manifestation
Field	From a control-only subfield toward a cross-domain space linking control, traffic, communication, computation, and infrastructure	Surveys now place modeling, communication, control, and system prospects in one frame; V2X and cloud-control work enters the core agenda 1 2
Ontology	From a linear vehicle string toward a heterogeneous vehicle-road-cloud-human coordination system	Mixed platoons, heterogeneous vehicle groups, human-in-the-loop digital twins, and coupled risk objects 12 13 14
Epistemology	From “model + stability proof” toward a hybrid evidence structure of proof, prediction, data, distillation, and platform validation	Lyapunov/MPC, Koopman-based predictive control, knowledge-guided learning, and digital-twin validation 3 6 8 9 13
Teleology	From maintaining spacing and string stability toward deployable safety, cooperative intelligence, eco-efficiency, and system-level risk governance	Eco-safe control, cooperative safety intelligence, risk entropy, and deployment-oriented validation 2 7 14

Start with the field layer. Vehicle platooning no longer circulates only inside control theory. It is being renamed inside a shared vocabulary of intelligent transportation, V2X, cloud coordination, cooperative perception, and CPS. In other words, the field’s location is shifting: researchers increasingly have to speak control, communication, sensing, computation, and deployment together rather than controller design alone.

Then consider ontology. The classical object was a line of vehicles and their longitudinal spacing. The newer object is much broader: heterogeneous vehicle groups, vehicle-road cooperative systems, cloud-coordinated group trajectories, and even human-vehicle-road risk couplings 12 13 14. The object is no longer simply “how a few vehicles follow each other,” but “how multiple actors coordinate under informational, control, and infrastructure constraints.”

The epistemic layer is shifting as well. Provability still matters, but analytical stability proof no longer monopolizes the truth position. What we now see is a hybrid evidence structure: control theory provides a normative center; data-driven prediction and knowledge distillation compensate for the limits of idealized models; and testbeds plus experimental platforms redefine what counts as credible evidence for deployment claims 6 8 9 13.

Teleology perhaps makes the expansion most visible. If the goal were only gap keeping and string stability, many peripheral materials could remain peripheral. But once the goal becomes deployable safety, cooperative intelligence, eco-efficiency, and system-level risk governance, then V2X, cloud control, risk entropy, human-in-the-loop settings, and digital twins stop looking supplementary and become part of the main line 2 7 14.

Why this layered reading is already a “University-style” frontier interpretation#

One thing should be made explicit: this field/ontology/epistemology/teleology map is itself already a surface scientific reading. Its strength is that it turns a complicated field into a map that experts can recognize, compare, and extend. Its risk is that it can temporarily freeze living contradictions and symptoms into neat classificatory boxes.

That is why I describe this step as a University-style frontier interpretation. It first organizes object, method, evidence, and goal into a teachable knowledge plane, allowing us to see how the field presents itself on the surface. But if we stop there, we still miss who sets the agenda, what cracks keep returning, and why new anchors repeatedly emerge. That is precisely why the next step has to move into a deeper four-discourse diagnosis.

Reading the field through Lacan’s four discourses#

If we only sort the literature by technical method, the field looks like a growing list of tools. But as an academic structure, vehicle platooning is better understood as something organized by four discourse positions.

Discourse	What it organizes	What it hides	What it produces
Master	Decides which problems matter most	High-level naming does not equal solvability	Legitimacy and direction
University	Classifies, models, and standardizes the object	The normative command behind the knowledge system	Papers, surveys, metrics, controller taxonomies
Hysteric	Forces existing answers to face real cracks	There is no final once-and-for-all answer	More questions and more knowledge production
Analyst	Reorganizes the field from its symptoms	A review is not a neutral mirror	New anchors and new problem statements

Master discourse: the high-level signifiers#

The dominant signifiers in vehicle platooning are easy to recognize: safety, string stability, cooperation, efficiency, zero-accident mobility, deployability. These are not merely goals. They also decide which questions are seen as inherently legitimate and urgent 2 3 4 7.

Once these signifiers occupy the center, later knowledge work naturally aligns itself around them. That is why papers with very different methods often begin with remarkably similar problem statements.

University discourse: the dominant mode of operation#

In practice, the dominant mode of operation in vehicle platooning is University discourse. The field works by classifying objects, standardizing problems, arranging methods into comparable matrices, and producing reviewable and teachable knowledge devices 1 5 6 16.

This is why so many papers share the same internal grammar: vehicle model, topology, delay setting, controller design, stability proof, simulation comparison. It is efficient and necessary, but it also tends to objectify the field’s living problems into scenarios and metrics.

Hysteric discourse: how the field keeps expanding#

What keeps the field from settling down is not knowledge’s smooth accumulation, but the persistence of unresolved remainders: what about communication delay, mixed traffic, heterogeneity, perception error, mismatch between cloud planning and actual trajectories, or real deployment 6 7 12 13 15?

These repeated “what about…” questions are precisely what keep forcing the field to produce more methods and more theory. Secondary literature is therefore not merely peripheral. It is often the product of the core being pressed by what it cannot easily absorb.

Analyst discourse: where a strong review should stand#

If a review only performs a better classification, it is still useful, but it remains inside University discourse. A stronger move is to ask why the field repeatedly circles back to safety, stability, and cooperation; why learning is repeatedly pulled back under engineering constraints; why testbeds suddenly matter so much; and why V2X, cloud control, and cooperative perception keep entering the same line of argument 2 14.

That is why I think the most valuable review position today is not simply to summarize what is known, but to show what drives the field, what it keeps hiding, and what it therefore keeps producing.

The four discourses are not static boxes but a rotating mechanism#

What still felt missing from the previous version of this article was not another classification table, but the dynamic rotation of the four discourses. Formally, Lacan’s four discourses are generated by quarter-turn movements of the same four elements—S1, S2, $, and a—through the same grid:

1
Master      S1 -> S2    / $  X a
2
University  S2 -> a     / S1 X $
3
Hysteric    $  -> S1    / a  X S2
4
Analyst     a  -> $     / S2 X S1

In the development of vehicle platooning as a research field, however, the most readable version of that movement is not a mechanical formula but a recurrent structural chain: master signifiers set the agenda → knowledge devices stabilize the object → real cracks force new demands → symptoms are reworked into a new anchor → the cycle begins again.

This should not be mistaken for a clean periodization, and it certainly does not mean that every paper completes a full turn. Different papers and subfields occupy different discourse positions simultaneously. Still, at the level of the field’s movement, this recurrent rotation is hard to miss.

Rotational moment	Structural action	How it appears in vehicle platooning
Master → University	High-level signifiers define the questions that must be answered	Safety, string stability, cooperation, and deployability pull methods into a common agenda 2 3 4 7
University → Hysteric	The knowledge apparatus meets remainders it cannot fully absorb	Delay, heterogeneity, mixed traffic, perception error, and deployment gaps keep returning 6 12 13 15
Hysteric → Analyst	Symptoms stop looking like edge cases and become entry points for reframing	V2X SPD, digital twins, and risk language begin to reorganize the problem itself 2 13 14
Analyst → new Master	A new anchor redefines what counts as the frontier problem	Vehicle platooning is rewritten as a networked CPS unit organized by communication, computation, and control 1 2 12 13 14

1) From Master to University: signifiers set the knowledge machine in motion#

When “safety,” “string stability,” “cooperation,” and “deployability” occupy the high ground, they do not solve anything directly. What they do first is organize the research collective. Surveys, monographs, controller taxonomies, evaluation metrics, and canonical scenarios are then built to answer those newly stabilized questions 1 3 4 16. In that sense, Master discourse sets the agenda, while University discourse translates that agenda into a knowledge system that can be modeled, proved, compared, and taught.

2) From University to Hysteric: the better the apparatus, the clearer the remainder#

But the more mature the knowledge apparatus becomes, the clearer its boundary becomes as well. Precisely because controllers, topologies, metrics, and validation routines are formalized so efficiently, the unabsorbed remainder keeps reappearing: human drivers in mixed traffic, intention errors under cooperative perception, degradation when delay crosses a boundary, trajectory mismatch in heterogeneous cloud-controlled groups, and the gap between “works in simulation” and “works in deployment” 4 6 12 13 15. This is the motor of Hysteric discourse: the field keeps asking authority to answer for the cracks it cannot eliminate.

3) From Hysteric to Analyst: symptoms become the entry point for reorganizing the problem#

If the field could only keep answering those “what about…” questions one by one, it would remain trapped in an ever-expanding University discourse of added models and patches. The turn happens when certain works start treating the cracks themselves as diagnostic clues: why do V2X, cooperative perception, cloud control, digital twins, and risk entropy rise together in the same period? That coincidence suggests that the object is no longer just a longitudinal control chain, but a system-level coupling across vehicles, roads, cloud coordination, humans, multi-source information, and validation platforms 2 13 14. In this sense, Analyst discourse does not offer one more refined taxonomy; it forces the field to admit that what once looked marginal is now rewriting the object itself.

4) From Analyst back to a new Master: how a new anchor starts the next cycle#

Once that reframing is accepted, a new S1 emerges. For the present field, that new anchor can be stated like this: vehicle platooning is not merely a longitudinal control problem, but a networked CPS unit organized by communication, computation, and control in coordination. This does not cancel the older priorities of safety and string stability. It relocates them within a larger systems frame: why cooperative perception matters, why edge/cloud coordination enters the discussion, why validation platforms become central method questions, and why risk modeling now belongs on the same line of inquiry 1 2 12 13 14. Once stabilized, that new S1 can in turn generate another round of University discourse: new taxonomies, new benchmarks, and new architectural vocabularies.

5) Why this rotation matters for review writing#

This is also why I no longer want the review to become a mere handbook of methods. If we only lay out controllers, learning algorithms, and testbeds side by side, the article remains mostly inside University discourse. A stronger review follows the rotational chain instead: first show which signifiers set the agenda, then show how knowledge devices stabilize the field, then show which real cracks forced the expansion, and finally show how those cracks produce a new anchor for the field. Written that way, a review no longer just summarizes knowledge. It shows how the field develops dynamically, reorganizes itself, and moves toward a networked CPS formulation.

What this means for review writing#

If the diagnosis above is roughly right—especially if the four discourses are understood as a dynamic rotation rather than as static boxes—then a contemporary review on vehicle platooning should not just become a long catalog of controllers. It should also explain:

how the field generates new frontiers through the cycle of agenda-setting, systematization, symptomatic pressure, reframing, and renewed agenda-setting;
why safety and string stability remain the long-term organizing center;
how communication delay, information topology, and heterogeneity turn a control problem into a networked control problem;
why data-driven, distilled, and reinforcement-based methods do not replace control theory but get pulled into its constraint structure;
why V2X, cloud control, cooperative perception, and experimental platforms push platooning toward a fuller CPS framing.

If I had to compress all of this into one sentence that could anchor a full review, it would be this:

Vehicle platooning research is moving from a longitudinal control problem toward a networked CPS problem organized by communication, computation, and control in coordination.

The value of this sentence is not that it replaces every existing taxonomy. Its value is that it places scattered materials back into one structure: why control theory remains central, why data-driven methods rise, why learning must be constrained, and why system validation has become a core issue.

Closing thought#

I no longer see vehicle platooning as a narrowly technical niche. It certainly began with car-following, stability, and control. But today it is becoming a very revealing intersection where control theory, traffic engineering, vehicular networking, edge/cloud coordination, cooperative perception, risk modeling, experimental platforms, and CPS language meet.

So the most interesting question is not just “what methods exist,” but how these methods are organized around a few dominant goals, and how they keep splitting, repairing, recombining, and generating new anchors under the pressure of real deployment gaps. For me, that is where the review becomes truly worth writing.

Reference Index#

Li, H., Meng, W., Han, Z., Zhang, Z., & Yang, Y. (2025). Vehicle platoon in road traffic: A survey of modeling, communication, controlling and perspectives. Physica A. DOI. li2025vehiclea
Zhang, J., Xu, Q., Li, Z., Xu, C., & Li, K. (2025). Cooperative safety intelligence in V2X-enabled transportation: A survey. arXiv. zhang2025cooperative
Ruan, T., Chen, Y., Li, X., Wang, J., Liu, Y., & Wang, H. (2025). Stability analysis and controller design of the cooperative adaptive cruise control platoon considering a rate-free time-varying communication delay and uncertainties. Transportation Research Part C. DOI. ruan2025stability
Tian, B., Guo, Q., Xu, Z., Wang, M., & Qu, X. (2026). MPC-based controller switching strategy for string stability and safety of vehicle platoons considering communication delays boundary. IEEE Transactions on Vehicular Technology. DOI. tian2026mpcbased
Li, Y., Liu, C., & Zheng, F. (2025). Robust fuzzy model predictive control for connected and automated vehicles in mixed platoons using a bidirectional vehicle dynamics strategy. Expert Systems with Applications. DOI. li2025robust
Li, S., Wang, J., Yang, K., Xu, Q., Wang, J., & Li, K. (2026). Robust nonlinear data-driven predictive control for mixed vehicle platoons via koopman operator and reachability analysis. Transportation Research Part C. DOI. li2026robust
Pan, X., Yuan, X., Ling, Z., Li, Z., Liu, K., & Wang, Y. (2026). A reinforcement learning-based decentralized control strategy for eco-safe mixed platooning with CAVs and HDVs. IEEE Transactions on Intelligent Transportation Systems. DOI. pan2026reinforcement
Wang, J., Wang, H., Song, J., Chen, X., Guo, J., Li, K., Li, X., & Huang, B. (2025). Knowledge-guided self-learning control strategy for mixed vehicle platoons with delays. Nature Communications. DOI. wang2025knowledgeguided
Wang, C., Jia, D., Wang, W., Ngoduy, D., Peng, B., & Wang, J. (2025). A knowledge-informed deep learning paradigm for generalizable and stability-optimized car-following models. Communications in Transportation Research. DOI. wang2025knowledgeinformed
Liu, G., Zheng, N., & Wang, H. (2025). Cooperative control method for connected and automated vehicle platoon based on arbitrary time headway switched system. SSRN. liu2025cooperative
Ruan, T., Chen, Y., Han, G., Wang, J., Li, X., Jiang, R., Wang, W., & Wang, H. (2025). Cooperative adaptive cruise platoon controller design considering switching control and stability. Transportation Research Part C. DOI. ruan2025cooperative
Shi, J., Guan, S., Zhong, W., Chen, C., Li, K., & Luo, Y. (2026). Robust predictive control of heterogeneous vehicle groups in the ramp cloud-control system using historical trajectory features. IEEE Transactions on Vehicular Technology. DOI. shi2026robust
Dong, J., Wang, J., Yang, C., Cai, M., Chen, C., Xu, Q., Wang, J., & Li, K. (2026). Multi-source human-in-the-loop digital twin testbed for connected and autonomous vehicles in mixed traffic flow. arXiv. dong2026multisource
Gao, H., Yang, H., Zhu, J., Su, H., Tang, C., Wang, X., Shi, J., Shen, C., & Liu, Z. (2026). Concept, principle, and modeling of driving risk entropy based on human-vehicle-road coupling model for autonomous vehicle. Neurocomputing. DOI. gao2026concept
Liu, K., Yuan, X., Li, Z., Pan, X., & Wang, Y. (2025). Robust driving intention prediction based on multi-stage learning under vehicle–infrastructure cooperative perception. IEEE Transactions on Intelligent Transportation Systems. DOI. liu2025robust
Xu, Z., Tian, B., & Zhang, Y. (2025). Optimization of Longitudinal Control for Truck Platoons Considering Communication Delays. China Science and Technology Publishing & Media. Link. xuzhigang2025kaolutongxinyanshidehuocheduiliezongxiangkongzhiyouhua