Cooperative Lane Changes for Connected and Automated Vehicles

Plan the lane changes, arrange the formation, and stitch the two motion stages. This MATLAB research implementation accompanies:

Bai Li, Youmin Zhang, Yiheng Feng, Yue Zhang, Yuming Ge, and Zhijiang Shao, “Balancing Computation Speed and Quality: A Decentralized Motion Planning Method for Cooperative Lane Changes of Connected and Automated Vehicles,” IEEE Transactions on Intelligent Vehicles, 3(3), 340–350, 2018. Paper / DOI.

What the example does

The example assigns initial and destination lanes to a group of vehicles, computes lane-change trajectories, adjusts their longitudinal spacing to avoid conflicts, and computes the preceding formation-adjustment stage. The two stages are then combined into a multi-vehicle maneuver.

The computation order differs from the driving order: Stage 2 is planned first because its required formation determines the target of Stage 1. The executed maneuver follows Stage 1 and then Stage 2.

flowchart LR
    A[Generate lane-change task] --> B[Stage 2: lane-change optimization]
    B --> C[Resolve conflicts by longitudinal translations]
    C --> D[Required initial formation for Stage 2]
    D --> E[Stage 1: formation adjustment]
    E --> F[Stitch Stage 1 then Stage 2]
    C --> F
    F --> G[Combined trajectories and visualization]

Loading

Requirements and quick start

• MATLAB on Windows. Some numerical and visualization components are distributed as MATLAB P-code.
• A usable AMPL installation and an AMPL-compatible IPOPT solver, with their matching runtime DLLs. Windows executables are included; third-party licensing and compatibility still apply.
• A writable repository directory: MATLAB and AMPL exchange local parameter, initialization, status, and trajectory files.

Open MATLAB in the repository root and run the entry point with its exact filename:

cd('C:/path/to/MultiCavLaneChange');
rng(1);                  % Optional: repeat the same randomized task
RunMe;

Keep the executables and DLLs together. If an external installation is used, make it visible to MATLAB's system commands, for example:

setenv('PATH', ['C:\path\to\ampl', pathsep, getenv('PATH')]);

The script leaves coop_traj_final in the workspace and calls the protected dsa.p renderer. tf1 and tf2 are stage durations; gross_time = tf1 + tf2 is the maneuver duration, not CPU time. Solver and graphics runtimes depend on the generated task and local installation.

The distribution is designed around its Windows executables and P-code; a native Linux/macOS workflow is not supplied here.

Configuration

Edit InitParams.m before running:

Parameter	Default	Meaning
params.user.num_cars	10	Number of cooperating vehicles
params.user.num_lanes	3	Available lanes
params.user.v_std	10 m/s	Nominal longitudinal speed
params.user.nfe_stage1	300	Stage 1 discretization points
params.user.nfe_stage2	100	Stage 2 discretization points
params.planner.ds	0.5 m	Longitudinal translation increment
params.vehicle.vmax	12 m/s	Speed bound
params.vehicle.amax	0.5 m/s²	Acceleration bound
params.vehicle.phymax	0.6 rad	Steering-angle bound
params.vehicle.wmax	0.25 rad/s	Steering-rate bound

The vehicle has a 2.8 m wheelbase, 0.96 m front overhang, 0.929 m rear overhang, and 1.942 m width. GenerateTask.m uses 3.75 m lane-center spacing and random lane assignments/gaps. Change that function when introducing a prescribed traffic scenario; changing the random seed alone does not prescribe the task.

Function map

File	Responsibility
RunMe.m	Orchestrate task generation, both optimizations, spacing adjustment, stitching, and visualization.
InitParams.m	Define vehicle, planner, and experiment settings in global params.
GenerateTask.m	Generate initial/destination lanes and write the task data file.
GenerateInitialGuessForStage2.m	Build initial lane-change state/control profiles.
WriteInitialGuessForStage2.p, WriteParamsForStage2.m	Prepare Stage 2 solver inputs.
SolveNlpAtStage2.p, LoadOptimumForStage2.p	Invoke and load the Stage 2 nonlinear optimization.
ProduceDelayList.m	Process vehicles in descending initial longitudinal position and translate conflicting trajectories in increments of ds.
IsCurTrajValidWithFixedPart.p	Test a candidate against already accepted vehicle trajectories.
WriteRelativeFormationInfoForStage1.m	Write the relative formation required by the adjusted Stage 2 trajectories.
GenerateInitialGuessForStage1.m	Build the formation-adjustment initialization.
WriteInitialGuessForStage1.p, WriteParamsForStage1.m	Prepare Stage 1 solver inputs.
SolveNlpAtStage1.p, LoadOptimumForStage1.p	Invoke and load the Stage 1 nonlinear optimization.
StitchTwoStages.p	Combine the two stages in execution order.
CreateVehiclePolygon.m	Compute the rectangular vehicle footprint from its rear-axle pose.
dsa.p	Protected visualization entry point called by RunMe.

Files ending in .p are executable MATLAB P-code, not missing .m files. The table describes their calling interfaces; their implementations are not distributed as readable source.

Citation

@article{Li2018CooperativeLaneChanges,
  author = {Bai Li and Youmin Zhang and Yiheng Feng and Yue Zhang
            and Yuming Ge and Zhijiang Shao},
  title = {Balancing Computation Speed and Quality: A Decentralized Motion
           Planning Method for Cooperative Lane Changes of Connected
           and Automated Vehicles},
  journal = {IEEE Transactions on Intelligent Vehicles},
  volume = {3}, number = {3}, pages = {340--350}, year = {2018},
  doi = {10.1109/TIV.2018.2843159}
}

License

See LICENSE for the repository's LGPL-2.1 terms. Bundled solvers and runtime libraries retain their own licenses.