Active Simulation-Based Inference for Scalable Car-Following Model Calibration

TL;DR

This work tackles the challenge of uncertainty-aware, driver-specific calibration of car-following models at scale. It proposes Active Simulation-Based Calibration (ASBC), which combines a residual-augmented IDM forward model with a trajectory-encoder-based neural posterior estimator (MAF) and an active joint parameter–scenario acquisition loop to produce driver-specific posteriors in a single forward pass at test time. Empirical results on the HighD dataset show that ASBC outperforms pooled Bayesian baselines in short-horizon predictive accuracy and distributional alignment, with the Matérn-$5/2$ residuals offering more temporally coherent uncertainty estimates. The framework enables scalable, uncertainty-aware driver population modeling for traffic simulation and risk-sensitive transportation analysis, while providing practical guidance and open problems for future extensions to black-box simulators and data-adaptive trajectory representations.

Abstract

Credible microscopic traffic simulation requires car-following models that capture both the average response and the substantial variability observed across drivers and situations. However, most data-driven calibrations remain deterministic, producing a single best-fit parameter vector and offering limited guidance for uncertainty-aware prediction, risk-sensitive evaluation, and population-level simulation. Bayesian calibration addresses this gap by inferring a posterior distribution over parameters, but per-trajectory sampling methods such as Markov chain Monte Carlo (MCMC) are computationally infeasible for modern large-scale naturalistic driving datasets. This paper proposes an active simulation-based inference framework for scalable car-following model calibration. The approach combines (i) a residual-augmented car-following simulator with two alternatives for the residual process and (ii) an amortized conditional density estimator that maps an observed leader--follower trajectory directly to a driver-specific posterior over model parameters with a single forward pass at test time. To reduce simulation cost during training, we introduce a joint active design strategy that selects informative parameter proposals together with representative driving contexts, focusing simulations where the current inference model is most uncertain while maintaining realism. Experiments on the HighD dataset show improved predictive accuracy and closer agreement between simulated and observed trajectory distributions relative to Bayesian calibration baselines, with convergence and ablation studies supporting the robustness of the proposed design choices. The framework enables scalable, uncertainty-aware driver population modeling for traffic flow simulation and risk-sensitive transportation analysis.

Figures (7)

• Figure 1: Classical per-trajectory Bayesian calibration versus ASBC. Left: obtain each driver’s posterior by running MCMC per trajectory; re-run when new data arrive. Right: train once with active joint parameter-scenario acquisition, then infer a driver-specific posterior by one forward pass.
• Figure 2: Amortized test-time inference module in ASBC. The training of this module is simulation-driven and relies on the active learning loop with joint parameter-scenario acquisition (Algorithm \ref{['alg:active_amortized_loop']}).
• Figure 3: Per-window 10-second simulation error distributions on HighD for ASBC. (a) ES and (b) RMSE over gap $s$ (m), speed $v$ (m/s), and acceleration $a$ (m/s$^2$). Colors indicate ASBC-Gaussian and ASBC-Matérn. Boxes show the interquartile range and the median.
• Figure 4: Case study on HighD pair #426. (a) ASBC-Gaussian and (b) ASBC-Matérn: posterior predictive rollouts with 95% PI (dashed: aligned leader speed). (c--d) Inferred posterior structure for the same pair, shown via pairwise marginals and the parameter correlation matrix.
• Figure 5: Active-loop convergence on HighD across rounds. The dashed line marks the round selected by the same validation-loss early-stopping rule used in training.