DRIVE: Dynamic Rule Inference and Verified Evaluation for Constraint-Aware Autonomous Driving

TL;DR

DRIVE introduces a unified framework that infers human-like soft driving constraints from demonstrations and integrates them into planning, achieving safe, compliant, and smooth autonomous driving trajectories with verified feasibility.

Contribution

It presents a novel probabilistic rule inference method combined with convex optimization-based planning for constraint-aware autonomous driving.

Findings

Achieves 0.0% soft constraint violation rate.

Produces smoother and more human-like trajectories.

Demonstrates strong generalization across diverse datasets.

Abstract

Understanding and adhering to soft constraints is essential for safe and socially compliant autonomous driving. However, such constraints are often implicit, context-dependent, and difficult to specify explicitly. In this work, we present DRIVE, a novel framework for Dynamic Rule Inference and Verified Evaluation that models and evaluates human-like driving constraints from expert demonstrations. DRIVE leverages exponential-family likelihood modeling to estimate the feasibility of state transitions, constructing a probabilistic representation of soft behavioral rules that vary across driving contexts. These learned rule distributions are then embedded into a convex optimization-based planning module, enabling the generation of trajectories that are not only dynamically feasible but also compliant with inferred human preferences. Unlike prior approaches that rely on fixed constraint forms or purely reward-based modeling, DRIVE offers a unified framework that tightly couples rule inference with trajectory-level decision-making. It supports both data-driven constraint generalization and principled feasibility verification. We validate DRIVE on large-scale naturalistic driving datasets, including inD, highD, and RoundD, and benchmark it against representative inverse constraint learning and planning baselines. Experimental results show that DRIVE achieves 0.0% soft constraint violation rates, smoother trajectories, and stronger generalization across diverse driving scenarios. Verified evaluations further demonstrate the efficiency, explanability, and robustness of the framework for real-world deployment.