Extensions to the Scenic Driving Domain

Scenic is a probabilistic programming language for generating and verifying environments of autonomous cyber-physical systems. Its driving domain- the primary interface for autonomous vehicle (AV) testing- is a Python library that parses OpenDRIVE road-network files and exposes lanes, roads, intersections, traffic signals, and agents as first-class Scenic objects. While the driving domain is powerful, it has several concrete limitations that hinder real-world usage: assertion failures and geometry artifacts on certain OpenDRIVE maps; a coarse-grained centerline-following model that accumulates error on curved roads; incomplete extraction of semantic road metadata; and absent lane-signal linkage that makes it impossible to write autopilot behaviors that respect traffic laws. This proposal addresses all of these gaps across two major tracks and five subtasks, matching the project outline provided by the mentors: - Track 1 (~4 weeks): Robustness - fix known bugs, expand the test suite. - Track 2 (~5 weeks): Features - Frenet-frame geometry, lane-following distance, semantic tags, signal-lane linkage, and modular scenario components. Each subtask explained in the proposal will produce independently reviewable pull requests, letting mentors integrate work incrementally. The result is a driving domain that is meaningfully more robust, more expressive, and better validated than today.