A critical review of motion planning methods for cooperative lane changes of connected and automated vehicles

Abstract

Cooperative lane-change motion planning coordinates the time-dependent lateral and longitudinal motions of a lane-changing vehicle with the anticipated or commanded motions of surrounding vehicles. Although related studies are often grouped by algorithms, traffic scenarios, or communication architectures, these dimensions overlap and do not provide mutually exclusive categories. This review focuses on methods that generate a continuous two-dimensional lane-change trajectory, or an equivalent sequence of lateral and longitudinal states or controls. A targeted Web of Science search with citation tracing retained 30 primary studies meeting this criterion. The studies are classified according to their primary planning output and the vehicles whose executable motions are directly determined. Four exclusive categories are established: interaction-conditioned ego-vehicle planning, fixed-structure cooperative motion planning, maneuver-organization-based multi-vehicle planning, and fully coupled multi-vehicle trajectory planning. The comparison shows that wider cooperative authority improves motion consistency and can resolve conflicts that are infeasible for an ego-only planner, but it also increases dependence on communication, behavioral models, computation, and centralized coordination. Across all four categories, uncertainty treatment is generally weaker than nominal optimization, safety constraints rarely constitute end-to-end guarantees, computational claims are difficult to compare, and validation remains dominated by designed simulations. Mixed traffic further changes cooperation from direct control to uncertain negotiation. These findings define the application conditions of the four planning paradigms and identify priorities for robust, scalable, and verifiable cooperative lane-change planning.

1 Introduction

Lane changing is a routine but safety-critical maneuver because it modifies the relative positions and safety margins of vehicles in both the original and target lanes. In dense traffic, the lane-changing vehicle cannot determine maneuver feasibility from its own trajectory alone. A feasible lateral transition may require the target-lane follower to decelerate, the target-lane leader to accelerate, several connected vehicles to adjust their spacing, or surrounding human drivers to yield. The planning problem is therefore coupled across vehicles, time, and lanes. Conventional autonomous-driving motion-planning surveys provide broad accounts of search, sampling, interpolation, optimization, prediction, and control methods (Paden et al., 2016; González et al., 2016; Schwarting et al., 2018; Claussmann et al., 2020), but cooperative lane changing poses a narrower question: how should interacting motions be converted into an executable two-dimensional maneuver?

Connectivity makes explicit cooperation possible by allowing vehicles or infrastructure to share states, intentions, candidate trajectories, and requested actions. Information exchange alone, however, does not guarantee cooperation. Vehicles may optimize incompatible objectives, a human driver may ignore an advisory, and a technically feasible joint plan may fail after delay, packet loss, or tracking error. Cooperative motion planning must consequently distinguish the availability of information from the authority to shape another vehicle’s motion. This distinction is especially important in mixed traffic, where other road users are observed and predicted but are not necessarily controllable. More broadly, energy and traffic-flow benefits attributed to connected and automated vehicles (CAVs) depend on the coordinated policy, penetration rate, and evaluation boundary rather than connectivity by itself (Vahidi and Sciarretta, 2018).

The term cooperative lane changing has also been used for substantially different technical objects. Some studies generate only a lane-change decision, a priority order, a target gap, or a longitudinal speed adjustment, while the lateral transition is assumed instantaneous or delegated to a traffic simulator. Other studies optimize a complete lateral-longitudinal trajectory for one vehicle while predicting the responses of others. A smaller group directly optimizes the continuous trajectories of several vehicles. Treating all of these studies as equivalent motion planners obscures the distinction between maneuver decision making and trajectory generation and makes method categories overlap.

This review therefore adopts a deliberately strict motion-planning boundary. A primary study is included in the four-category synthesis only when its planner generates a time-parameterized two-dimensional lane-change trajectory or an equivalent sequence of lateral and longitudinal states or controls. The trajectory may be optimized jointly, selected from a parameterized family, or generated by a hierarchical planner, but it must represent the lateral transition explicitly. Studies limited to lane-change approval, merge order, longitudinal gap preparation, or instantaneous lane transitions are outside the core synthesis. They are adjacent coordination or traffic-control studies rather than cooperative lane-change motion planners under the definition used here.

The contribution of this review is threefold. First, it defines the task and its boundaries in terms of planning output rather than terminology used by individual authors. Second, it proposes four mutually exclusive categories based on cooperative decision authority and motion coupling. Third, it compares the categories through traffic scenario, vehicle composition, planning architecture, uncertainty treatment, safety constraints, computational efficiency, validation method, advantages, limitations, and application conditions. The objective is not to identify one universally superior planner. It is to explain what each paradigm can coordinate, which assumptions make it effective, and where its claims cease to be reliable.

The remainder of the paper is organized as follows. Section 2 defines the scope, search process, and study-selection rules. Section 3 positions this review relative to existing surveys. Section 4 presents the exclusive four-level taxonomy. Sections 5–8 critically examine the four categories. Section 9 discusses cross-category trade-offs, practical barriers, research gaps, and mixed traffic involving CAVs and human-driven vehicles (HDVs). Section 10 concludes the review.

2 Review scope and methodology

2.1 Operational definition and scenario boundary

In this review, cooperative lane-change motion planning denotes the generation of a time-dependent lateral-longitudinal motion for a lane-changing vehicle whose feasibility, safety, or objective explicitly depends on the coordinated or anticipated motion of at least one other traffic participant. Cooperation may be explicit, when connected vehicles or infrastructure exchange intentions and determine compatible executable motions, or implicit, when the ego planner models another driver’s response and adapts its trajectory accordingly. These two forms are not treated as equivalent. Explicit cooperation provides greater decision authority but depends on connectivity and compliance, whereas implicit cooperation remains conditional on behavioral prediction.

The scope includes discretionary and mandatory lane changes on multilane roads, platoon lane changes, lane drops, dedicated-lane entry, and weaving or ramp-merging maneuvers when the planner explicitly represents the continuous lateral transition. A merging study is therefore included only when it solves a lane-stream transition with a two-dimensional motion output; longitudinal arrival-time control at a fixed merge point is not sufficient. Similarly, mixed traffic is included when human-driver behavior is incorporated into a two-dimensional automated vehicle (AV) planner, but ordinary traffic-flow simulation without an explicit planning output is not.

Three exclusion boundaries are applied. First, lane-change decision models that output only change, keep, yield, or priority actions are excluded unless a downstream module generates a two-dimensional trajectory. Second, longitudinal gap-opening or speed-harmonization methods are excluded when lateral motion is omitted, instantaneous, or supplied entirely by a microscopic simulator. Third, studies concerned only with tracking a given path, cooperative perception, communication protocols, or general traffic management are excluded because they do not solve the lane-change motion-planning problem itself. These boundaries reduce the breadth of the corpus but improve conceptual consistency.

2.2 Literature-search and selection procedure

The review corpus was reconstructed through a targeted structured search of the Web of Science Core Collection. The publication window extended from January 2010 through July 2026, and the search was last updated on 1 August 2026. The topic query combined terms for the lateral maneuver, cooperation, and motion generation, including model predictive control (MPC). Using the Web of Science Topic field, the search syntax was: TS=((“lane chang*” OR “lane-change*” OR merg* OR weaving) AND (cooperat* OR collaborat* OR “connected automated vehicle*” OR CAV* OR “multi-vehicle” OR “multi-agent”) AND (“motion planning” OR “trajectory planning” OR “trajectory generation” OR “trajectory optimization” OR “maneuver planning” OR “optimal control” OR “model predictive control” OR MPC)). The database search was supplemented by backward and forward citation tracing from the eligible papers and from reviews of autonomous-driving motion planning, lane-changing models, freeway merging, and cooperative driving. This supplementary step was necessary because several relevant papers describe a cooperative maneuver through terms such as distributed optimization, platoon coordination, or interaction-aware planning without placing “cooperative lane change” in the title.

Eligible records were English-language journal articles or peer-reviewed conference papers that: (i) addressed a lane change or an equivalent lateral lane-stream transition; (ii) modeled interaction with at least one surrounding vehicle; (iii) generated a continuous two-dimensional trajectory or an equivalent lateral-longitudinal state or control sequence; and (iv) reported sufficient methodological and validation information for critical assessment. Records were excluded if they provided only a decision, priority, target gap, longitudinal controller, instantaneous lane transition, or simulator-internal lane-change execution. General surveys and conceptual papers were retained only for background and novelty comparison and were not assigned to a method category.

The search and citation tracing produced a working pool of 43 technical papers that addressed cooperative or interaction-dependent lane changing, merging, weaving, or equivalent lane-stream transitions. Full texts were then checked against one decisive planning-output criterion. Thirteen papers were removed because they determined only whether or when to change lanes, a merge or priority order, a target gap, a longitudinal gap-opening action, or a lane transition executed internally by a traffic simulator. These papers are relevant to cooperative decision making or traffic control, but they do not generate the complete two-dimensional motion reviewed here. The remaining 30 primary studies were retained and assigned to exactly one of the four method categories. Ten surveys or broad background papers were used only to define terminology, establish review novelty, and position the technical corpus; they were not counted as primary methods. Because this is a focused Mini Review rather than a bibliometric or systematic-review claim, unrecoverable initial-hit and duplicate counts are not reported.

2.3 Data extraction and critical appraisal

Each primary study was examined using a common set of methodological and evidential attributes, including traffic scenario, vehicle composition, directly controlled vehicles, lateral and longitudinal planning variables, planning architecture, treatment of interaction and uncertainty, safety formulation, computation time or update rate, and validation method. For concise cross-study comparison, these attributes were consolidated into five dimensions in Tables 1–4: Scenario and Participants, Planning Architecture, Interaction and Uncertainty, Safety Treatment, and Computation and Validation. Category-defining information, particularly the vehicles whose executable motions are directly determined, was assessed separately according to the classification rules in Section 4.

Validation evidence was distinguished among designed numerical simulation, microscopic traffic simulation, naturalistic-data replay, vehicle-dynamics simulation, hardware-in-the-loop (HIL) testing, and physical-vehicle experiments. These forms of evidence were not treated as equivalent. For example, replaying recorded surrounding trajectories does not provide closed-loop interaction evidence because the recorded vehicles cannot respond to a counterfactual ego maneuver.

The appraisal also distinguished safety-related mechanisms from safety guarantees. Time headway, time to collision, potential fields, collision penalties, and sampled geometric constraints may improve nominal safety but do not by themselves establish robustness to prediction error, actuation error, communication failure, or between-sample collision. Likewise, computation time was interpreted together with the reported hardware, planning horizon, vehicle count, algorithmic scope, and required update interval rather than as an isolated indicator of real-time capability. Category assignment was based on full-text methodological reading using the explicit rules in Section 4. Keyword co-occurrence or bibliometric clustering was not used to construct the taxonomy because terminology alone does not determine which vehicles receive executable motions.

3 Positioning relative to existing reviews

Broad reviews of autonomous-driving motion planning organize methods by algorithmic family, planning hierarchy, driving function, or road environment. Paden et al. (2016) surveyed planning and control for self-driving urban vehicles, González et al. (2016) compared motion-planning techniques for AVs, and Claussmann et al. (2020) focused on highway motion planning. Schwarting et al. (2018) connected decision making, prediction, planning, verification, and multi-agent interaction. These works establish the general methodological foundation, but cooperative lane changing occupies only a small part of their scope and is not classified according to the multi-vehicle motion that the planner directly determines.

Lane-change reviews address a closer behavior but pursue a different analytical objective. Zheng (2014) distinguished lane-change decision and lane-change impact models and emphasized behavioral calibration and traffic-flow effects. Ma and Li (2023) reviewed lane-change decision and trajectory-prediction research from a broad bibliometric and modeling perspective. These surveys reveal important shortcomings in driver modeling and validation, but they do not focus on how connected or interaction-aware planners convert cooperation into executable multi-vehicle motion.

Reviews of merging and cooperative driving provide complementary system perspectives. Zhu et al. (2022) classified freeway on-ramp control by traffic composition and control setting and covered sequence, trajectory, game, learning, and feedback methods. Häfner et al. (2022) examined cooperative architectures, maneuver negotiation, intent exchange, and request-response protocols. Ji et al. (2024) reviewed cooperative vehicle-infrastructure systems, including communication, roadside perception, edge computing, datasets, and applications. These reviews explain where information and coordination authority originate, but they do not use continuous two-dimensional motion generation as an inclusion boundary or compare planner-level safety, uncertainty, computation, and validation on a common basis.

The novelty of the present review is therefore not simply that it discusses lane changing. It narrows the object of analysis to cooperative motion planners with explicit lateral-longitudinal outputs, applies a single exclusive classification rule, and critically compares the evidence needed for implementation. The resulting taxonomy is organized by what a planner decides and for whom it decides, rather than by traffic scenario or solver label. This distinction prevents MPC, game theory, reinforcement learning, or distributed computation from being mistaken for mutually exclusive classes.

4 Four-level taxonomy and classification rules

The classification axis is the primary planning output together with the set of vehicles whose executable motions are directly determined. The four categories form a progression from conditional ego planning to fixed cooperative structures, maneuver-level organization, and fully coupled multi-vehicle trajectory optimization.

Interaction-conditioned ego-vehicle planning generates only the ego vehicle’s executable trajectory. Surrounding vehicles influence the plan through predicted reactions, latent intentions, or strategic response models, but their trajectories are not commanded by the planner. Fixed-structure cooperative motion planning directly coordinates multiple vehicles while keeping their roles, insertion relation, or formation membership predetermined. Its principal simplification is to regulate spacing, timing, or reference tracking within that fixed structure. Maneuver-organization-based multi-vehicle planning makes the organization itself a decision variable, including passing order, target slot, grouping, coalition, priority, or action sequence, and then generates a two-dimensional motion under the selected organization. Fully coupled multi-vehicle trajectory planning directly optimizes the continuous state or control trajectories of several vehicles in a common coupled problem.

Assignment follows the dominant methodological output rather than every module contained in a study. If only the ego receives an executable motion and all other vehicles remain response models, the study belongs to the first category. If several vehicles are directly coordinated but their roles and maneuver structure are fixed and motion is generated through reduced, staged, or reference-based planning, it belongs to the second. If the central contribution determines order, grouping, coalition, or another maneuver structure and continuous trajectory generation is subordinate to that result, it belongs to the third. If complete multi-vehicle trajectories are the primary coupled decision variables, the study belongs to the fourth even when the cooperative group or insertion order is predefined. Solver choice, communication topology, traffic composition, and uncertainty model are recorded as comparison dimensions rather than alternative categories.

This progression concerns cooperative authority and motion coupling, not a guaranteed ordering of numerical runtime. A discrete ordering problem in the third category can be computationally harder than a small continuous program in the fourth. Nevertheless, the taxonomy provides an increasing resolution of the cooperative planning object: response-conditioned ego motion, reduced motion within a fixed structure, organization plus downstream motion, and direct optimization of the joint multi-vehicle motion.

5 Interaction-conditioned ego-vehicle planning

Interaction-conditioned planners are positioned at the boundary between conventional ego-centric planning and explicit connected cooperation. Their executable output is an ego trajectory, but that trajectory is conditioned on how surrounding vehicles are expected to respond to a merge request, a narrowing gap, or an assertive lateral motion. This approach is particularly relevant when the surrounding vehicles are human-driven or cannot accept direct control commands. Its central technical difficulty is not multi-vehicle actuation but counterfactual prediction: the planner must estimate how another vehicle would react to each candidate ego action.

5.1 Decision-conditioned prediction and adaptive ego planning

Sheng et al. (2023) combined candidate reference decisions, a graph-based spatiotemporal predictor, and nonlinear MPC. Predictions were conditioned on the ego candidate rather than generated independently, which reduced the inconsistency between behavior prediction and motion planning. The final MPC optimized acceleration and steering while enforcing road, actuation, and geometric collision constraints against the selected predictions. The method demonstrates an important advantage of interaction-conditioned planning: a lane change that appears infeasible under constant-velocity prediction may become feasible when surrounding traffic is expected to respond. The same advantage is also its main vulnerability. Point predictions are embedded as if they were reliable future motions, so distribution shift or an unexpectedly non-yielding vehicle can invalidate the nominal safety constraint.

Burger et al. (2020) represented the target-lane follower through cooperative and noncooperative intention models. A Bayesian estimator updated the intention probabilities, and an interaction-aware mixed-integer quadratic program (MIQP) generated a shared near-term ego motion followed by intention-dependent branches. Although the optimization contains multi-agent trajectories, only the ego plan is executable; the follower branches describe possible responses. The method therefore belongs in this category rather than in fully coupled planning. Its branching formulation avoids an abrupt commitment to one behavioral label, but the binary intention set, preset transition parameters, and small three-vehicle scenario limit generality. No computation time or large-scale validation was reported.

Liu et al. (2023) developed a leader-follower game controller for forced merging. The ego selected among precomputed lane-keeping, lane-change, and abort trajectories, while Bayesian inference estimated whether target-lane vehicles would proceed or yield. Chance constraints incorporated model-conditional uncertainty, and tests included 198 Next-Generation Simulation (NGSIM) merge cases. This study provides stronger evidence than a few designed simulations and explicitly includes an abort option. However, NGSIM replay is not closed-loop real-traffic validation because recorded vehicles cannot react to a counterfactual ego trajectory. The pairwise decomposition and binary leader-follower roles also simplify multi-vehicle dependence, and the reported average computation time in the NGSIM cases was close to the stated update-period limit.

5.2 Strategic response models and driving-style adaptation

Wu et al. (2024) combined driving-style clustering, polynomial ego candidates, and a Stackelberg game with the target-lane follower. The equilibrium trajectory was subsequently tracked through an adaptive MPC framework. Naturalistic data were used to estimate styles and assess decision and trajectory prediction, while CarSim-Simulink experiments examined tracking under different speeds and road adhesion. The end-to-end connection from response modeling to vehicle-dynamics tracking is a strength. Nevertheless, the style classes and payoff weights are coarse, the follower remains a modeled responder rather than a connected participant, and safety is mainly expressed through headway rewards rather than a hard robust constraint.

Qin et al. (2026) expanded strategic interaction to several surrounding HDVs. Intelligent Driver Model (IDM) parameters and fuzzy logic were used to identify driving style, a level-k mixed-strategy game represented lane-change and yield choices, and clothoid-based path planning with dynamic programming (DP) and quadratic programming (QP) speed optimization produced the ego trajectory. The method includes non-yielding responses and abort behavior and was demonstrated on a physical vehicle platform. Its breadth is valuable, but it also introduces many model layers, weights, thresholds, and identification assumptions. A few designed experiments cannot establish the reliability of the full pipeline under broader driver diversity, and the safety logic remains model-dependent.

5.3 Advantages, limitations, and application conditions

The principal advantage of this category is deployability in mixed or weakly connected traffic. It does not require surrounding vehicles to expose their control variables or accept a centralized plan. Decision-conditioned prediction also prevents the ego planner from treating human behavior as independent of its own maneuver. These methods are therefore appropriate for dense traffic in which the ego must negotiate for space but cannot command the other vehicle.

The limitation is that predicted cooperation is not actual cooperation. The ego trajectory can indirectly encourage yielding, but feasibility depends on a behavioral model whose validity is difficult to certify. Binary intentions and fixed driving-style classes simplify inference but underrepresent ambiguous, time-varying, or adversarial behavior. More elaborate games and learned predictors improve expressiveness while adding calibration, computational, and distribution-shift risks. Application should therefore be conditional on conservative fallback behavior, calibrated uncertainty, explicit abort trajectories, and an independent safety layer. Without these elements, interaction awareness may reduce conservatism at the cost of overconfidence.

6 Fixed-structure cooperative motion planning

Fixed-structure planners directly coordinate two or more connected vehicles, but the cooperating roles and maneuver relation are determined before trajectory generation. A subject vehicle may request a particular target-lane follower to decelerate, a fixed triplet may prepare an insertion gap, or an intact platoon may change lanes under a leader-follower topology. By restricting the cooperative structure, these planners obtain smaller and more interpretable problems than general joint planning. The cost is reduced adaptability when the selected participant, target gap, or reference profile becomes unsuitable.

6.1 Candidate-based and parameterized cooperation

Bai et al. (2017) generated quintic longitudinal and lateral candidates for an accelerated lane change and allowed the target-lane follower to maintain speed or decelerate. Candidate trajectories were filtered by kinematic, comfort, and rectangular-vehicle collision constraints, and the cost explicitly included the burden imposed on the follower. The approach is transparent and geometrically more faithful than point-mass gap checking, but it assumes that the requested collaborator follows one of a few deterministic modes and does not replan under communication or behavioral failure.

Zong et al. (2022) used a potential-field module to choose the lane-change opportunity, generated polynomial trajectories for the subject and participating vehicles, and invoked cooperation when an unassisted lane change was infeasible. The method includes noncooperative manual traffic and abort or recovery behavior in microscopic simulation. Its candidate set and cost weights nevertheless remain heuristic, and the human response is deterministic rather than probabilistic. Wei et al. (2024) enlarged the candidate space by allowing longitudinal and lateral actions to start and end asynchronously. Continuous collision-space analysis identified feasible parameter regions, which improved flexibility relative to synchronized polynomials. The method reported sub-0.1-s online computation for selected stages, but the result depends on discretization, precomputation, exact states, and a restricted polynomial family.

These three methods illustrate the main benefit of parameterization: safety and comfort checks can be conducted rapidly over a manageable set of interpretable motions. Their limitation is representational. A trajectory outside the selected polynomial or timing family is unavailable even when dynamically feasible, and enlarging the family directly increases search cost. They are best suited to local interactions with a small number of candidate collaborators and sufficiently predictable surrounding motion.

6.2 Staged gap preparation and lateral execution

Chen et al. (2022) separated cooperative gap preparation from lateral execution. A connected vehicle and two target-lane vehicles first adjusted their longitudinal states under time and energy objectives; lateral motion was then generated using optimal-control, parametric, or control barrier function (CBF) alternatives. This separation yields a tractable problem and clarifies the cost of cooperation, but it weakens the representation of lateral-longitudinal coupling during the transition. Sun et al. (2021) adopted a related reduction by prescribing lateral motion and using two longitudinal stages with constant accelerations. The method included disturbance to followers in both lanes and screened several maneuver durations, but equal phase lengths, constant accelerations, and a small number of designed cases restrict fidelity.

Wang et al. (2024) applied a two-stage structure to freeway on-ramp lane changing. A fixed triplet first optimized longitudinal headway adjustment with engine-delay dynamics and a variable duration; a quintic lateral reference and MPC tracking controller then executed the lateral transition. Incorporating vehicle dynamics and ramp-length constraints improves physical relevance, but the longitudinal and lateral stages remain decoupled and the lateral stage assumes constant longitudinal speed. The comparison with the linear quadratic regulator (LQR) primarily evaluates tracking rather than the complete cooperative planner.

Wang et al. (2016) coordinated the lane-changing vehicle and target-lane neighbors with longitudinal MPC while prescribing the lateral acceleration as a sinusoidal function. Large initial-condition sweeps showed that cooperation increased modeled lane-change feasibility and reduced upstream disturbances. This is useful traffic-level evidence, but homogeneous connected traffic, a deterministic car-following model, intuitive weights, and the absence of communication or tracking uncertainty limit transferability. Ni et al. (2020) also combined an incentive assessment with a two-stage MPC and a prescribed sinusoidal lateral profile. Naturalistic trajectories were used to initialize selected simulations, but the surrounding traffic was modified counterfactually and therefore did not provide closed-loop real-world evidence.

Sun et al. (2023) addressed the switching of the longitudinal predecessor during lane changing. A rolling MPC smoothly transferred safety and tracking emphasis as the lane-changing vehicle crossed lane boundaries. The method reported computation below a 0.1-s cycle in two designed CAV scenarios and reduced abrupt acceleration relative to a noncooperative baseline. Its fixed lateral curve, exact-state assumption, and very small scenario set make the result conditional rather than general.

6.3 Fixed platoon and facilitator structures

Nie et al. (2022) treated a heterogeneous platoon as an intact moving formation. The leader planned a collision-free lane-change trajectory and followers tracked spacing-based references through distributed model predictive control (DMPC) under several communication topologies. Heterogeneous parameters and distributed control broaden the model beyond identical-vehicle platoons, and stability conditions were provided. However, the leader’s fine-discretization trajectory optimization required several seconds, the target lane was largely unobstructed, and delay or packet loss was not tested despite the emphasis on communication topology.

Li et al. (2018) first transformed a CAV group into a sparse longitudinal formation and then executed simultaneous lane changes. Exact rectangular collision checks and a reduced formation-regulation problem provided a deliberate computation-quality compromise, supported by randomized trials and completeness reasoning under the model assumptions. The approach can require substantial road length and several seconds of preparation, and its guarantees depend on fully controllable CAVs, exact execution, and adequate space. It is therefore more appropriate for planned group maneuvers than for sudden dense-traffic conflicts.

Kim and Borrelli (2023) assigned the leading CAV as a facilitator. The facilitator entered the target lane, regulated a receiving gap, and enabled the remaining CAVs to follow through DMPC and finite-state machine (FSM) coordination. Randomized scenes initialized from naturalistic data showed substantially more successful maneuvers than a simultaneous-lane-change baseline, but the absolute success rate remained limited. Facilitator selection was predetermined, other traffic was not allowed to change lanes, and expected or soft safety constraints did not establish a probabilistic guarantee.

6.4 Advantages, limitations, and application conditions

Fixed-structure cooperation offers the clearest balance between online tractability and direct multi-vehicle coordination. The number of controlled vehicles is small or organized by a known topology, objectives are interpretable, and MPC or polynomial searches can often operate at practical update rates. The category is suitable when the target gap and cooperating vehicles can be identified reliably, such as connected triplets, scheduled platoons, or infrastructure-managed road segments.

Its central weakness is structural rigidity. A planner may optimize motions precisely while assuming the wrong collaborator, gap, order, or formation. Most methods also simplify the lateral transition through a prescribed curve, staged execution, or fixed maneuver duration. This reduction limits adaptation when longitudinal and lateral constraints become simultaneously active. Communication failures, refusal to cooperate, and tracking errors are rarely incorporated into the feasibility analysis. Consequently, these planners require participant-selection logic and a fallback mechanism outside the fixed optimization problem.

7 Maneuver-organization-based multi-vehicle planning

Maneuver-organization-based planners determine how several vehicles should share constrained road space before or together with trajectory generation. Their defining variables include passing order, target gap, vehicle group, coalition structure, priority, or action sequence. Unlike fixed-structure planners, they do not assume that the relevant organization is known. Unlike fully coupled trajectory planners, they reduce the continuous problem by solving an organizational layer and then generating motions under the selected structure.

7.1 Order, slot, and group organization

Xu et al. (2020) decomposed cooperative driving in conflict areas into an upper-level passing-order search and a lower-level trajectory planner. Monte Carlo tree search (MCTS) and heuristics explored orders, while a predefined trajectory set generated feasible motions for the selected order. Randomized experiments showed a favorable compromise between first-in-first-out (FIFO) computation and exhaustive-search quality for small groups. The separation is interpretable and scalable relative to enumeration, but near-optimality depends on search budget and the lower-level trajectory library. Full control authority, reliable communication, and known states are assumed.

Li et al. (2020) grouped CAVs according to conflict relationships before jointly optimizing polynomial trajectories within each group. Earlier groups constrained later groups, reducing the dimension of several simultaneous mandatory lane changes. Grouping makes a problem with many vehicles manageable, but the maximum group size and grouping distance materially affect completion, oscillation, and solution quality. Because these parameters are prescribed rather than optimized, the decomposition can sacrifice global optimality without a quantitative bound. Computation time was also not reported despite the online framing.

Jiang et al. (2024) used DP to assign mandatory lane-changing CAVs to ordered target-lane gaps. A longitudinal MPC assessed candidate costs and a dynamically refitted cubic lateral curve executed the selected maneuver. The method is notable for combining sequence assignment with a two-dimensional downstream plan and for evaluating mixed traffic across several demands with repeated microscopic simulations. Its HDVs nevertheless follow fixed behavioral models, the prediction is deterministic, and the mandatory fallback can force a merge when no satisfactory plan has been found. The cubic path also imposes a fixed-duration kinematic structure rather than solving a fully coupled trajectory.

7.2 Coalition, priority, and learned organization

Hang et al. (2022) formulated multi-lane merging decisions as a coalitional game among CAVs with different preference weights. Candidate coalitions and discrete actions were evaluated through predictive vehicle dynamics, and individual rationality was represented using Shapley-value concepts. The method exposes a real tension between system-optimal coordination and whether each participant benefits sufficiently to join. However, coalition structures are evaluated for only small groups, perfect cloud and vehicle-to-vehicle (V2V) information is assumed, and driving styles are assigned objective weights rather than uncertain inferred behaviors. Scalability beyond the demonstrated three- and five-vehicle cases remains unresolved.

Alshakhs et al. (2025) used a Markov decision process (MDP), conflict detection, dynamic priority, an S-shaped lateral trajectory, and proportional-integral-derivative (PID)-based longitudinal control for coordinated multi-lane changes in mixed traffic. The method offers a complete decision-to-motion pipeline and resolves overlapping lane-change requests through priority. Its state and action abstractions, fixed maneuver duration, deterministic human behavior, and limited safety model make it more appropriate as a high-level coordination prototype than as a vehicle-dynamics planner. The reported experiments also do not establish performance under communication impairment or broad traffic distributions.

Zhang et al. (2024) learned cooperative action policies with multi-agent reinforcement learning (MARL) and introduced prioritized action extrapolation so lower-priority vehicles could account for higher-priority predicted actions. Parameter sharing and global information enabled several connected vehicles with different target lanes to coordinate in two- and three-lane simulations. The method handles changing priorities without explicit combinatorial enumeration, but collisions were reduced rather than eliminated, actions were coarse, and evaluation used only a small number of random tests after extensive training. No hard safety filter, timing result, mixed-HDV transfer, or real deployment evidence was provided.

7.3 Advantages, limitations, and application conditions

The strength of maneuver organization is that it addresses conflicts a continuous optimizer should not be forced to discover blindly. Choosing a reasonable order, slot, group, or coalition can reduce dimensionality and prevent simultaneous incompatible requests. These planners are particularly useful in bottlenecks, mandatory lane-change zones, multi-vehicle target-lane assignments, and situations with several competing CAV intents.

The weakness is the interface between organization and motion. A high-level order can be efficient but dynamically infeasible, while a lower-level planner may be too restricted to repair it. Hierarchical methods therefore require explicit feasibility feedback rather than one-way command transfer. Priority and utility rules can also encode arbitrary weights, unfairness, or local decisions that are not system-optimal. Learning-based organization adds flexibility but shifts the problem toward training coverage, safety certification, and generalization. These methods are most credible when the lower-level trajectory generator is included in candidate evaluation and when organizational choices can be revised online.

8 Fully coupled multi-vehicle trajectory planning

Fully coupled planners place the continuous states or controls of several vehicles in a common optimization problem. Cooperation is not represented only through a requested gap or an upper-level order; it emerges from simultaneous satisfaction of shared collision, road, timing, formation, and dynamic constraints. This formulation provides the richest motion consistency and the clearest representation of reciprocal effects. It also produces the most severe growth in variables, nonconvex collision constraints, mixed discrete-continuous choices, communication requirements, and sensitivity to model mismatch.

8.1 Centralized formulations

Burger and Lauer (2018) formulated cooperative multi-vehicle planning as a centralized MIQP. Longitudinal and lateral motions, road boundaries, obstacles, and pairwise collision avoidance were represented in a unified problem. MIQP provides a structured treatment of discrete conflict choices and continuous trajectories, but binary variables and simplified dynamics restrict scale and physical fidelity. The work demonstrates the value of a common joint configuration space rather than a deployment-ready online planner.

Li et al. (2017) formulated a full multi-vehicle optimal control problem (OCP) with bicycle kinematics, collision avoidance, road constraints, and completion-time and steering objectives. Progressively Constrained Dynamic Optimization (PCDO) activated only collision constraints found to be necessary, making a difficult 12-vehicle problem solvable. The online implementation first regulated vehicles into a standard formation and then used offline optimized patterns. This decomposition is technically effective but shifts substantial computation offline, can require seconds of online preparation, and creates a lookup-table scaling problem as vehicle and lane counts grow.

Luo et al. (2019) jointly optimized the continuous trajectories of two simultaneous lane-changing vehicles under same-direction and intersecting-direction conflict patterns. Geometric safety spacing and rectangular collision relationships captured interactions ignored by simple longitudinal gaps, and CarSim or HIL elements connected planning to vehicle dynamics. The planner was nevertheless evaluated offline, surrounding vehicles maintained constant speed, and scaling beyond two active lane changers was proposed rather than demonstrated.

Duan et al. (2023) unified multi-vehicle and platoon motion planning in a nonlinear program with Ackermann kinematics, formation requirements, road constraints, and pairwise geometric collision avoidance. The method can represent both formation preservation and reconfiguration and provides a useful high-quality benchmark. The authors appropriately position it as computation-intensive rather than real-time. Evidence is limited to two designed seven-vehicle cases, and the locally solved deterministic problem contains no margin for communication, perception, model, or tracking error.

Liu et al. (2024) combined an extended A* (E-A*) initialization with convex-feasible-set-based sequential convex programming (CFS-SCP) for minimum-time multi-vehicle lane-change planning. Repeated perturbed-start tests, runtime comparisons, and HIL implementation provide stronger algorithmic evidence than isolated demonstrations. However, the upper-level optimization still required several seconds, collision constraints were enforced at discretized nodes, and the separate deep reinforcement learning (DRL) response to an unexpected obstacle did not establish robustness of the complete planner. The controlled ground-vehicle setting also omits human traffic interaction.

8.2 Distributed and decomposed formulations

Tajalli et al. (2022) embedded continuous longitudinal-lateral polynomial trajectories and binary lane-change decisions in a distributed receding-horizon mixed-integer nonlinear programming (MINLP) formulation for freeway segments with lane drops. Iterative vehicle communication approximated a system-level solution and enabled traffic-scale VISSIM evaluation. The study is unusually informative about computational breakdown: the update target was met at lower demands but exceeded under saturated conditions even on high-performance hardware. Perfect communication, homogeneous CAV traffic, fixed lane-change duration, Big-M safety constraints, and a weak congested baseline limit the magnitude and transferability of the reported network gains.

Katriniok (2020) decomposed nonlinear Frenet-model trajectory optimization using the consensus alternating direction method of multipliers (ADMM). Each vehicle solved a private local problem, while a subject-vehicle coordinator enforced shared headway consistency. The method retained lateral and longitudinal dynamics, limited exchanged information, and reported detailed per-iteration timing that was mostly below a 0.1-s cycle after initialization. The result is promising for small connected groups, but the nonconvex algorithm has no global convergence guarantee, the first initialization exceeded the cycle, and validation consisted of one nominal three-CAV scenario with assumed low latency and no packet loss.

8.3 Advantages, limitations, and application conditions

The primary advantage of fully coupled planning is internal motion consistency. Collision relationships, dynamic limits, lane boundaries, shared timing, and participant costs can be evaluated within one mathematical object rather than passed imperfectly between modules. This category is most appropriate for tightly controlled CAV groups, benchmark generation, infrastructure-managed zones, and situations where simultaneous lateral conflicts cannot be represented by fixed gap logic.

Its practical limitations are substantial. Pairwise collision constraints grow rapidly, nonconvex programs yield local solutions, mixed-integer formulations scale poorly, and distributed methods require repeated communication within each planning cycle. A high-quality nominal trajectory may be obsolete before it is computed. More importantly, optimizing several vehicles jointly assumes that the planned participants are controllable and will track the communicated trajectory with sufficient accuracy. Without robust margins, delay-aware coordination, and a certified fallback, the apparent consistency of the centralized plan can conceal fragile implementation assumptions.

9 Discussion

9.1 Principal trade-offs across the four categories

The four categories reveal a trade-off between cooperative authority and assumption burden. Interaction-conditioned ego planners require the least direct authority and can operate around human drivers, but they inherit uncertainty from response prediction. Fixed-structure planners control more vehicles and can create a feasible gap deliberately, but they depend on selecting the correct participants and maneuver relation before optimization. Maneuver-organization planners resolve competing intents and combinatorial conflicts, but their high-level result must remain compatible with downstream dynamics. Fully coupled planners provide the strongest internal consistency, but require the largest shared state, computation, communication, and control authority.

A second trade-off concerns motion freedom and tractability. Parameterized polynomials, sinusoidal lateral profiles, fixed durations, staged gap preparation, and reference tracking reduce computation and often make real-time implementation plausible. These reductions are not merely weaknesses; they are design choices that can be appropriate in structured highway maneuvers. Their limitation appears when feasible motion requires simultaneous adaptation of lateral timing, longitudinal acceleration, target gap, and participant roles. Fully coupled formulations restore that freedom but encounter nonconvexity and scaling. The relevant comparison is therefore not simplistic optimality versus heuristics, but whether the chosen reduction removes motion needed in the intended operating domain.

Centralized and distributed planning create another nontrivial trade-off. Centralized planners can evaluate system-level objectives and reciprocal constraints consistently, while distributed planners preserve local computation, model privacy, and some scalability. Distributed optimization does not eliminate coordination cost; it replaces a large program with local solves and iterative communication. Its performance depends on initialization, message latency, iteration limits, and the behavior of nonconvex consensus. Conversely, centralized formulations may be acceptable for a small infrastructure-managed zone but become a single point of failure in open traffic.

Finally, individual maneuver success is not equivalent to traffic-level benefit. A target-lane follower may sacrifice comfort, a sparse formation may consume road length, and a conservative safety headway may move congestion upstream. Some studies measure local time, jerk, or feasibility, while others report network speed or delay. These outcomes cannot be compared directly. Evaluation should state the spatial and temporal boundary of each metric and separate ego benefit, participant cost, and system effect.

9.2 Practical barriers and strength of current evidence

Communication is usually modeled as perfect state and intent exchange. A few studies account for an assumed message latency or compare communication topologies, but systematic treatment of packet loss, stale plans, clock mismatch, corrupted messages, or participant dropout is rare. This omission is most serious for fully coupled and distributed planners because the safety constraint itself may depend on a trajectory received from another vehicle. Future evaluation should include explicit delay and loss models, asynchronous updates, bounded inconsistency, and a transition from cooperative to independent safe control when communication degrades.

Model uncertainty is also treated unevenly. Some interaction-conditioned methods estimate latent intentions or use model-conditional chance constraints, whereas many CAV-only methods assume exact dynamics and perfect tracking. A label such as uncertainty-aware should not be used when uncertainty consists only of randomized initial states or a single unexpected obstacle. Robustness must specify the uncertain quantity, its assumed set or distribution, how it enters prediction and constraints, and whether the guarantee survives model mismatch.

Safety terminology requires particular restraint. Geometric collision constraints, braking-distance rules, time headways, potential fields, collision penalties, and episode termination provide different levels of assurance. A sampled nominal collision constraint does not guarantee continuous-time separation, and a probabilistic constraint is meaningful only under its distributional assumptions. None of the reviewed paradigms provides a complete guarantee covering perception error, human response, communication failure, numerical approximation, and tracking error. Claims should therefore distinguish nominal collision-free planning, model-conditional probabilistic safety, recursive feasibility, controller-level invariance, and end-to-end operational safety.

Computational results are difficult to compare because studies report different layers and hardware. Some reported times exclude prediction, participant selection, communication, or initialization; others report only one nominal scenario. A credible real-time claim should report end-to-end latency distributions over varying vehicle counts and densities, identify the required cycle, include worst-case or high-percentile values, and describe the fallback when the deadline is missed. The reviewed evidence shows that local fixed-structure planners can often approach subsecond execution, whereas large joint formulations may require seconds or specialized hardware. However, no universal runtime ranking follows from category alone.

Validation remains dominated by simulation. Naturalistic-data initialization improves scenario realism but does not validate closed-loop interaction. Recorded-data replay cannot test how humans respond to a new AV action. Vehicle-dynamics simulation and HIL testing address actuation and tracking but may still prescribe surrounding traffic. Only a small number of studies use physical vehicles, and these demonstrations cover limited cases. A stronger validation ladder should progress from reproducible simulation and randomized stress testing to driver-in-the-loop studies, communication and HIL testing, controlled proving-ground trials, and ultimately supervised road experiments.

9.3 Research controversies and future directions

One controversy is the meaning of cooperation. A planner that predicts a human driver’s response may be interaction-aware without being cooperative in the connected-control sense. Conversely, a connected planner may share information without optimizing a common or mutually acceptable objective. Future work should report the cooperation mechanism explicitly: information sharing, negotiated intent, advisory influence, direct control, or joint optimization. This vocabulary would prevent behavioral prediction and commanded multi-CAV coordination from being compared as if they provided the same authority.

A second controversy concerns planning hierarchy. Separating organization, trajectory generation, and tracking improves modularity, but one-way interfaces can produce infeasible or unsafe plans. Fully integrated optimization improves consistency but can be too slow or opaque. A promising direction is bidirectional hierarchy: the motion layer returns feasibility regions, risk bounds, and computation estimates to the organizational layer, which revises order or roles before committing. Such an architecture can preserve interpretability without assuming that a high-level decision is always executable.

Further development should prioritize scalable uncertainty-aware coordination rather than simply adding more objective terms. Useful directions include set- or distributionally robust interaction models, communication-aware distributed optimization, safety filters independent of learned policies, adaptive participant selection, anytime planning with certified intermediate solutions, and standardized benchmarks spanning vehicle count, density, penetration rate, and communication quality. Reporting practices also require improvement. Source code, scenario definitions, complete timing, unsuccessful cases, baseline tuning, and confidence intervals are necessary for meaningful comparison.

9.4 Mixed CAV-HDV traffic as a problem of limited cooperation

Mixed traffic is not a fifth planning category because vehicle composition cuts across all four mechanisms. It changes the meaning of cooperation by reducing control authority. A connected CAV can receive an executable trajectory, an instrumented HDV may receive advice but choose whether to follow it, and an unconnected human driver can only be observed and predicted. These participants should not be represented by the same cooperative constraint.

Several retained studies already expose this transition. Interaction-conditioned planners estimate yielding, proceeding, driving style, or strategic response without assuming direct communication (Sheng et al., 2023; Burger et al., 2020; Liu et al., 2023; Wu et al., 2024; Qin et al., 2026). Other methods embed simplified HDV models or noncooperative traffic around connected participants (Zong et al., 2022; Kim and Borrelli, 2023; Jiang et al., 2024; Alshakhs et al., 2025). These studies show that mixed traffic can be represented, but current models often reduce humans to IDM parameters, fixed style classes, constant-speed prediction, or scripted responses. Such representations are useful for controlled comparison but insufficient for a safety claim.

The central mixed-traffic question is therefore not merely CAV penetration rate. It is whether a planner remains safe when an assumed collaborator is unobservable, unwilling, delayed, or inconsistent. Future planners should assign each participant an explicit cooperation status and confidence level, update that status online, and reserve a maneuver only when the residual risk is acceptable. Advice acceptance and inferred yielding should be probabilistic, while refusal or communication loss should trigger a recoverable fallback rather than an infeasible joint plan. Evaluation should include heterogeneous vehicle dynamics, noncompliant connected vehicles, human response diversity, and interaction experiments in which surrounding drivers can react to the AV in closed loop.

10 Conclusion

This review examined cooperative lane-change motion planning under a strict continuous two-dimensional motion criterion and organized 30 studies into four categories. Interaction-conditioned ego planners adapt to predicted responses without directly controlling other vehicles. Fixed-structure planners coordinate known participants under a predetermined maneuver relation. Maneuver-organization planners determine order, slot, group, coalition, or priority before producing motion. Fully coupled planners directly optimize continuous trajectories for several vehicles.

The categories should be viewed as application-dependent alternatives rather than stages in a universal performance ranking. Increasing cooperative authority can resolve tighter conflicts and improve motion consistency, but it also increases dependence on communication, controllability, model accuracy, computation, and reliable execution. The reviewed literature spans an expanding range from response-aware ego motion to direct joint multi-vehicle planning, but current evidence does not support a general claim that the field has completed a shift toward fully organized interaction. The most persistent gaps are robust treatment of communication and behavior uncertainty, scalable computation, precise safety claims, fair cross-method evaluation, and closed-loop real-world validation. Progress on these issues is necessary before cooperative lane-change planners can move from favorable simulations to dependable mixed-traffic operation.

References

• 1

  AlshakhsA.MysorewalaM.NasirA. (2025). Coordinated multi lane-changing on highways for connected and automated vehicles in mixed traffic. Expert Syst. Appl.286, 127890. 10.1016/j.eswa.2025.127890

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  BaiH.ShenJ.WeiL.FengZ. (2017). Accelerated lane-changing trajectory planning of automated vehicles with vehicle-to-vehicle collaboration. J. Adv. Transp.2017, 8132769. 10.1155/2017/8132769

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  BurgerC.LauerM. (2018). “Cooperative multiple vehicle trajectory planning using MIQP,” in 2018 21st IEEE International Conference on Intelligent Transportation Systems (ITSC), 602–607. 10.1109/ITSC.2018.8569776

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  BurgerC.SchneiderT.LauerM. (2020). “Interaction aware cooperative trajectory planning for lane change maneuvers in dense traffic,” in 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC), 1–8. 10.1109/ITSC45102.2020.9294638

  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  ChenR.CassandrasC. G.Tahmasbi-SarvestaniA.SaigusaS.Nourkhiz MahjoubH.Al-NadawiY. K. (2022). Cooperative time and energy-optimal lane change maneuvers for connected automated vehicles. IEEE Trans. Intell. Transp. Syst.23, 3445–3460. 10.1109/TITS.2020.3036420

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  ClaussmannL.RevilloudM.GruyerD.GlaserS. (2020). A review of motion planning for highway autonomous driving. IEEE Trans. Intell. Transp. Syst.21, 1826–1848. 10.1109/TITS.2019.2913998

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  DuanX.SunC.TianD.ZhouJ.CaoD. (2023). Cooperative lane-change motion planning for connected and automated vehicle platoons in multi-lane scenarios. IEEE Trans. Intell. Transp. Syst.24, 7073–7091. 10.1109/TITS.2023.3253479

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  GonzálezD.PérezJ.MilanésV.NashashibiF. (2016). A review of motion planning techniques for automated vehicles. IEEE Trans. Intell. Transp. Syst.17, 1135–1145. 10.1109/TITS.2015.2498841

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  HäfnerB.BajpaiV.OttJ.SchmittG. A. (2022). A survey on cooperative architectures and maneuvers for connected and automated vehicles. IEEE Commun. Surv. Tutor.24, 380–403. 10.1109/COMST.2021.3138275

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  HangP.LvC.HuangC.XingY.HuZ. (2022). Cooperative decision making of connected automated vehicles at multi-lane merging zone: a coalitional game approach. IEEE Trans. Intell. Transp. Syst.23, 3829–3841. 10.1109/TITS.2021.3069463

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  JiY.ZhouZ.YangZ.HuangY.ZhangY.ZhangW.et al (2024). Toward autonomous vehicles: a survey on cooperative vehicle-infrastructure system. iScience27, 109751. 10.1016/j.isci.2024.109751

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  JiangY.ManZ.WangY.YaoZ. (2024). Cooperative lane-changing for connected autonomous vehicles merging into dedicated lanes in mixed traffic flow. Expert Syst. Appl.252, 124163. 10.1016/j.eswa.2024.124163

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  KatriniokA. (2020). Nonconvex consensus ADMM for cooperative Llane change maneuvers of connected automated vehicles. IFAC-PapersOnLine53, 14336–14343. 10.1016/j.ifacol.2020.12.1379

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  KimH.BorrelliF. (2023). Facilitating cooperative and distributed multi-vehicle lane change maneuvers. IFAC-PapersOnLine56, 5585–5590. 10.1016/j.ifacol.2023.10.463

  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  LiB.ZhangY.GeY.ShaoZ.LiP. (2017). 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), in Optimal Control-based Online Motion Planning for Cooperative Lane Changes of Connected and Automated Vehicles, 3689–3694. 10.1109/IROS.2017.8206215

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  LiB.ZhangY.FengY.ZhangY.GeY.ShaoZ. (2018). Balancing computation speed and quality: a decentralized motion planning method for cooperative lane changes of connected and automated vehicles. IEEE Trans. Intell. Veh.3, 340–350. 10.1109/TIV.2018.2843159

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  LiT.WuJ.ChanC.-Y.LiuM.ZhuC.LuW.et al (2020). A cooperative lane change model for connected and automated vehicles. IEEE Access8, 54940–54951. 10.1109/ACCESS.2020.2981169

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  LiuK.LiN.TsengH. E.KolmanovskyI.GirardA. (2023). Interaction-aware trajectory prediction and planning for autonomous vehicles in forced merge scenarios. IEEE Trans. Intell. Transp. Syst.24, 474–488. 10.1109/TITS.2022.3216792

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  LiuT.ChaiR.ChaiS.ArvinF.ZhangJ.LennoxB. (2024). Fast collision-free multi-vehicle lane change motion planning and control framework in uncertain environments. IEEE Trans. Ind. Electron.71, 16602–16613. 10.1109/TIE.2024.3398674

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  LuoY.YangG.XuM.QinZ.LiK. (2019). Cooperative lane-change maneuver for multiple automated vehicles on a highway. Automot. Innov.2, 157–168. 10.1007/s42154-019-00073-1

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  MaC.LiD. (2023). A review of vehicle lane change research. Phys. A626, 129060. 10.1016/j.physa.2023.129060

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  NiJ.HanJ.DongF. (2020). Multivehicle cooperative lane change control strategy for intelligent connected vehicle. J. Adv. Transp.2020, 8672928. 10.1155/2020/8672928

  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  NieG.XieB.LuH.TianY. (2022). A cooperative lane change approach for heterogeneous platoons under different communication topologies. IET Intell. Transp. Syst.16, 53–70. 10.1049/itr2.12128

  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  PadenB.CápM.YongS. Z.YershovD.FrazzoliE. (2016). A survey of motion planning and control techniques for self-driving urban vehicles. IEEE Trans. Intell. Veh.1, 33–55. 10.1109/TIV.2016.2578706

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  QinZ.WangH.ChenL.HuangC.BianY. (2026). Lane-changing decision-making and planning based on multi-vehicle game interaction. IEEE Trans. Veh. Technol.75, 3786–3799. 10.1109/TVT.2025.3612454

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  SchwartingW.Alonso-MoraJ.RusD. (2018). Planning and decision-making for autonomous vehicles. Annu. Rev. Control Robot. Auton. Syst.1, 187–210. 10.1146/annurev-control-060117-105157

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  ShengZ.LiuL.XueS.ZhaoD.JiangM.LiD. (2023). A cooperation-aware lane change method for automated vehicles. IEEE Trans. Intell. Transp. Syst.24, 3236–3251. 10.1109/TITS.2022.3225875

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  SunK.ZhaoX.WuX. (2021). A cooperative lane change model for connected and autonomous vehicles on two lanes highway by considering the traffic efficiency on both lanes. Transp. Res. Interdiscip. Perspect.9, 100310. 10.1016/j.trip.2021.100310

  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  SunK.ZhaoX.GongS.WuX. (2023). A cooperative lane change control strategy for connected and automated vehicles by considering preceding vehicle switching. Appl. Sci.13, 2193. 10.3390/app13042193

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  TajalliM.NiroumandR.HajbabaieA. (2022). Distributed cooperative trajectory and lane changing optimization of connected automated vehicles: freeway segments with lane drop. Transp. Res. Part C Emerg. Technol.143, 103761. 10.1016/j.trc.2022.103761

  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  VahidiA.SciarrettaA. (2018). Energy saving potentials of connected and automated vehicles. Transp. Res. Part C Emerg. Technol.95, 822–843. 10.1016/j.trc.2018.09.001

  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  WangD.HuM.WangY.WangJ.QinH.BianY. (2016). Model predictive control-based cooperative lane change strategy for improving traffic flow. Adv. Mech. Eng.8, 1–17. 10.1177/1687814016632992

  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  WangZ.XiangJ.WangJ.GaoZ.ChenT.LiH.et al (2024). Cooperative lane-changing control for CAVs at freeway on-ramps considering vehicle dynamics. J. Adv. Transp.2024, 1221717. 10.1155/2024/1221717

  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  WeiL.ZhangW.BaiH.LiJ. (2024). 2D-action asynchronous cooperative lane change trajectory planning method for connected and automated vehicles. J. Adv. Transp.2024, 5540444. 10.1155/2024/5540444

  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  WuD.LiZ.DuC.LiuC.LiY.XuX. (2024). Lane changing control of autonomous vehicle with integrated trajectory planning based on stackelberg game. IEEE Open J. Intell. Transp. Syst.5, 810–825. 10.1109/OJITS.2024.3509462

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  XuH.ZhangY.CassandrasC. G.LiL.FengS. (2020). A bi-level cooperative driving strategy allowing lane changes. Transp. Res. Part C Emerg. Technol.120, 102773. 10.1016/j.trc.2020.102773

  • CrossRef
  • Google Scholar
  • View reference in article

• 37

  ZhangX.LiS.WangB.XueM.LiZ.LiuH. (2024). “Multi-vehicle collaborative lane changing based on multi-agent reinforcement learning,” in 2024 IEEE Intelligent Vehicles Symposium (IV), 1214–1221. 10.1109/IV55156.2024.10588529

  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  ZhengZ. (2014). Recent developments and research needs in modeling lane changing. Transp. Res. Part B Methodol.60, 16–32. 10.1016/j.trb.2013.11.009

  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  ZhuJ.EasaS.GaoK. (2022). Merging control strategies of connected and autonomous vehicles at freeway on-ramps: a comprehensive review. J. Intell. Connect. Veh.5, 99–111. 10.1108/JICV-02-2022-0005

  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  ZongF.HeZ.ZengM.LiuY. (2022). Dynamic lane changing trajectory planning for CAV: a multi-agent model with path preplanning. Transp. B Transp. Dyn.10, 266–292. 10.1080/21680566.2021.1989079

  • CrossRef
  • Google Scholar
  • View reference in article