Connected Vehicle Supported Adaptive Traffic Control for Near-congested Condition in a Mixed Traffic Stream
Pith reviewed 2026-05-24 21:55 UTC · model grok-4.3
The pith
Adaptive traffic signals using only 5% connected vehicle data raise speeds and cut queues in near-congested mixed traffic.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The central claim is that a connected-vehicle-supported adaptive signal control algorithm, which forecasts total platoon traffic counts from CV data via machine learning and then applies multi-objective optimization to determine green intervals plus real-time offset adjustment, improves major-street operational conditions under near-congested mixed traffic relative to loop-detector-based actuated coordinated control, with the gains observable at 5% CV penetration.
What carries the argument
Machine learning prediction of total traffic counts from CV platoons combined with multi-objective optimization of green intervals and dynamic offset adjustment.
If this is right
- Benefits to both CV and non-CV vehicles increase as the fraction of connected vehicles rises.
- The same CV-only data stream can support real-time adaptation without additional detector infrastructure.
- Operational improvements are achieved specifically in the coordinated direction of the major street during near-congested periods.
Where Pith is reading between the lines
- Cities could phase out loop detectors on arterials if the prediction accuracy holds in live traffic.
- The platoon-based optimization might extend to networks with multiple coordinated arterials.
- Real-world driver behavior differences could alter the observed queue and delay reductions.
Load-bearing premise
The machine learning model trained on CV data accurately predicts the total number of vehicles in each platoon, and the simulation faithfully reproduces real mixed-traffic behavior.
What would settle it
Field measurements on the same corridor showing that the prediction error exceeds ten vehicles or that the reported speed, queue, and delay improvements do not appear when 5% CV data are supplied to the algorithm.
read the original abstract
Connected Vehicles (CVs) have the potential to significantly increase the safety, mobility, and environmental benefits of transportation applications. In this research, we have developed a real time adaptive traffic signal control algorithm that utilizes only CV data to compute the signal timing parameters for an urban arterial in the near congested condition. We have used a machine learning based short term traffic forecasting model to predict the overall traffic counts in CV based platoons. Using a multi objective optimization technique, we compute the green interval time for each intersection using CV based platoons. Later, we dynamically adjust intersection offsets in real time, so the vehicles in the major street can experience improved operational conditions compared to loop detector based actuated coordinated signal control. Using a 3 mile long simulated corridor of US 29 in Greenville, SC, we have evaluated the performance of our CV based adaptive signal control. For the next time interval, using only 5% CV data, the Root Mean Square Error of the machine learning based prediction is 10 vehicles. Our analysis reveals that the CV based adaptive signal control improves operational conditions in the major street compared to the actuated coordinated scenario. Also, using only CV data, the operational performance improves even for a low CV penetration (5% CV), and the benefit increases with increasing CV penetration. We can provide operational benefits to both CVs and non CVs with the limited data from 5% CVs, with 5.6% average speed increase, and 66.7% and 32.4% reduction in average maximum queue length and stopped delay, respectively, in major street coordinated direction compared to the actuated coordinated scenario in the same direction.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper develops a real-time adaptive traffic signal control algorithm for near-congested urban arterials that relies exclusively on connected vehicle (CV) data. A machine-learning short-term forecasting model predicts total (CV + non-CV) platoon counts from 5 % CV penetration; these predictions feed a multi-objective optimizer that sets green intervals, after which offsets are adjusted dynamically. The system is tested in VISSIM on a 3-mile segment of US 29 in Greenville, SC, and is reported to outperform loop-detector-based actuated coordinated control, delivering a 5.6 % increase in average speed and 66.7 % / 32.4 % reductions in maximum queue length and stopped delay (major-street coordinated direction) even at 5 % CV penetration. The abstract states an RMSE of 10 vehicles for the platoon-count predictor.
Significance. If the quantitative claims hold after the prediction-error scale is clarified, the work would show that usable adaptive control is achievable with minimal CV market penetration and without additional detector infrastructure, benefiting both equipped and unequipped vehicles. The simulation-based evaluation on a real corridor geometry supplies a concrete, falsifiable benchmark that future field studies could test.
major comments (2)
- [Abstract] Abstract: the reported RMSE of 10 vehicles for platoon-count prediction is given without any accompanying mean, median, or distribution of the platoon volumes being predicted. Because the central performance gains (5.6 % speed, 66.7 % queue, 32.4 % delay) are produced by feeding these predictions into the multi-objective optimizer, the absence of scale context leaves it impossible to judge whether a 10-vehicle error is negligible or decision-altering.
- [Results / Evaluation section] Results / Evaluation section: the reported operational improvements are presented as single-point estimates with no error bars, no sensitivity analysis on the ML hyperparameters or the multi-objective weights, and no description of how the forecasting model was trained or cross-validated. These omissions directly affect the load-bearing claim that the CV-based controller outperforms the actuated baseline at low penetration.
minor comments (1)
- [Abstract] Abstract: the sentence 'We can provide operational benefits to both CVs and non CVs with the limited data from 5% CVs' is slightly redundant with the preceding clause and could be tightened for clarity.
Simulated Author's Rebuttal
We thank the referee for the constructive comments, which highlight opportunities to strengthen the clarity and rigor of our work. We address each major comment below and commit to revisions that directly respond to the concerns raised.
read point-by-point responses
-
Referee: [Abstract] Abstract: the reported RMSE of 10 vehicles for platoon-count prediction is given without any accompanying mean, median, or distribution of the platoon volumes being predicted. Because the central performance gains (5.6 % speed, 66.7 % queue, 32.4 % delay) are produced by feeding these predictions into the multi-objective optimizer, the absence of scale context leaves it impossible to judge whether a 10-vehicle error is negligible or decision-altering.
Authors: We agree that context on platoon volume scale is necessary to interpret the RMSE. In the revised manuscript we will add the mean, median, and distribution (or range) of the predicted platoon volumes from our VISSIM simulations to both the abstract and the methods/results sections. This will allow readers to assess the relative magnitude of the 10-vehicle error. revision: yes
-
Referee: [Results / Evaluation section] Results / Evaluation section: the reported operational improvements are presented as single-point estimates with no error bars, no sensitivity analysis on the ML hyperparameters or the multi-objective weights, and no description of how the forecasting model was trained or cross-validated. These omissions directly affect the load-bearing claim that the CV-based controller outperforms the actuated baseline at low penetration.
Authors: We acknowledge that the current manuscript presents point estimates without accompanying variability measures or sensitivity checks and omits details on model training. The revised version will incorporate error bars from multiple simulation replications, a sensitivity analysis on ML hyperparameters and multi-objective weights, and an explicit description of the forecasting model's training and cross-validation procedure. These additions will better substantiate the performance claims at low CV penetration. revision: yes
Circularity Check
No circularity; claims rest on independent simulation evaluation of ML prediction and optimization.
full rationale
The derivation chain consists of an ML model trained to predict total platoon counts from partial CV data, followed by multi-objective optimization of green times and real-time offset adjustment, with performance quantified via VISSIM simulation outputs (speed, queue, delay metrics). No step reduces by the paper's own equations to a fitted parameter renamed as a prediction, nor to a self-citation that is itself unverified; the reported RMSE=10 and percentage improvements are simulation-derived quantities, not tautological with the inputs. The evaluation remains externally falsifiable against the actuated baseline within the same simulator.
Axiom & Free-Parameter Ledger
free parameters (2)
- ML forecasting model parameters
- Multi-objective optimization weights
axioms (2)
- domain assumption CV observations at low penetration suffice to forecast total traffic counts
- domain assumption The VISSIM model of the Greenville corridor reproduces real near-congested mixed traffic behavior
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.