GLOSA Trajectory Optimization

This repository contains research code and supporting notebooks for Green Light Optimal Speed Advisory (GLOSA) trajectory optimization. The methods compute energy-efficient vehicle trajectories for signalized intersections under deterministic or uncertain traffic-signal switching times.

1. Vehicle Trajectory Specification in Presence of Traffic Lights with Known or Uncertain Switching Times
   Panagiotis Typaldos, Ioanna Kalogianni, Kyriakos Simon Mountakis, Ioannis Papamichail, Markos Papageorgiou. Transportation Research Record (2020), 53-66.

2. Modified dynamic programming algorithms for GLOSA systems with stochastic signal switching times
   Panagiotis Typaldos, Markos Papageorgiou. Transportation Research Part C, 157 (2023), 104364.

The code focuses on deterministic and stochastic trajectory generation for vehicles approaching signalized intersections, with implementations based on Stochastic Dynamic Programming (SDP), Discrete Differential Dynamic Programming (DDDP), and Differential Dynamic Programming (DDP).

Status

This repository contains research code associated with published work. It is maintained as a reference implementation rather than a production software package.

Research focus

The repository studies how an approaching vehicle should shape its trajectory when the next traffic-light switching time is:

• known in advance, as in fixed-time signal plans;
• uncertain but described by a switching-time distribution;
• solved with faster approximate dynamic-programming variants suitable for more demanding online use cases.

Each major source file corresponds to a concrete algorithmic experiment or paper-era comparison.

Repository map

red2green/
  dp/
    main.c                   Baseline dynamic programming implementation
  dddp/
    *.c
    mathematica/             Early stochastic and iterative DP notebooks
  ddp/
    *.nb                     Mathematica notebooks for DDP-related analysis
  journal TRC/
    sdp.c
    dddp.c
    ddp.c
    coefficients.h
    mathematica/             Notebooks associated with the TRC paper

all signal phases/
  MFTS22/
    mfts.cpp
    final_times.c
    plots.py
    signal_example.png
    *.txt                    Precomputed tables and sample outputs for this experiment set
    pet/                     Companion variant of the same experiment family

Where to start

• red2green/dp/ is the clearest starting point for the baseline stochastic DP formulation.
• red2green/journal TRC/ is the most publication-ready code area for the 2023 TRC paper and contains the SDP, DDDP, and DDP comparison implementations.
• red2green/dddp/ and red2green/ddp/ retain earlier development notebooks and supporting implementations that help trace the research progression.
• all signal phases/MFTS22/ contains a separate experiment family with supporting precomputed tables, plotting scripts, and an illustrative signal figure.

Building representative programs

The programs are standalone research executables configured primarily through constants and scenario arrays near the top of each source file.

Build with make

make

This creates:

• build/glosa_dp
• build/glosa_sdp
• build/glosa_dddp
• build/glosa_ddp

You can remove generated binaries with:

make clean

Baseline DP example

./build/glosa_dp

TRC paper examples

./build/glosa_sdp
./build/glosa_dddp
./build/glosa_ddp

Some programs write output files into folders such as outputs/ or results_sdp/. These directories are treated as generated artifacts and are intentionally excluded from version control. When a particular experiment expects one of those folders, create it before running the executable.

glosa_dddp first looks for a saved SDP initialization trajectory. If that file is not present, it reports the missing path and automatically falls back to the built-in Scenario 1 initial trajectory so the executable remains runnable.

Data and generated artifacts

The large j_opt_*.txt and te_opt_*.txt files under all signal phases/MFTS22/ are retained because the accompanying C++ experiments read them as precomputed inputs. Other files named like output_*.txt are treated as run outputs rather than source assets.

Notes for reproducibility

• Scenario definitions, initial conditions, and mode switches are currently encoded in the source files rather than exposed through a unified command-line interface.
• Several notebooks document derivations, plots, and experiment analysis alongside the C/C++ implementations.
• The repository has been kept close to the paper-era folder layout so that code, notebooks, and publication artifacts remain easy to cross-reference.

Publications and citation

For citation support on GitHub, see CITATION.cff.