Autonomous racing: models and data

Policy interfaces and synthetic loader checks

All 15 evaluation policies passed the directed synthetic interface comparisons. A candidate terminal-filter correction matched an independent oracle on 16,380 synthetic dtype cases; the original helper failed 4,094.

F1TENTH research car from the simulator documentation. Platform reference photograph.Credit: Joseph Auckley, Matthew O’Kelly, Aman Sinha and Hongrui Zheng. MIT. Image reproduced unchanged.Full-size image Image source

results and contributions

Policy interfaces checked	15 on directed synthetic inputs
Candidate correction / oracle agreement	16,380 of 16,380 synthetic dtype cases
Original helper failures	4,094 on the same finite domain

Repository Evidence at source commit Current roadmap Author’s canonical benchmark Corrected-source qualification Exhaustive synthetic filter result

figures and hardware

Open any image at full size. Source links identify the retained revision.

Pinned archive structure: recording segments, policy-name strings and terminal-flag alignment. Neither names nor flags prove independent executable policy episodes. — **Archive qualification.** Historical archive census at the original pins. The registrations at those pins contained archive aliases; this figure is not a census of the corrected releases. Recording segments and policy names do not establish independent executable episodes.Full-size image Image source

**Public-data development comparison.** Retained command-response comparison on licensed driving data. It does not identify tire properties or validate transfer between vehicles.Full-size image Image source

design evolution

The design and analysis steps that produced the current work.

Work	Engineering question	Approach	Outcome
Simulator controllers	Identification used telemetry generated by the same model being fitted.	Retain it as a simulation consistency exercise.	Implemented pure pursuit, LQR and MPC for simulator comparisons.
LiDAR mount	A 285.5 Hz mode is crossed by motor shaft order near 8.3 m/s.	Reopen the mounting requirement.	Calculated the shaft-order crossing near 8.3 m/s, identifying a limitation of the original 200 Hz design criterion.

methods and scope

These are source-code and synthetic-interface results. The candidate correction has not been adopted on archives. Corrected ZIP qualification, reuse terms and the intended simulator revision precede empirical controller rankings. Historical archive figures retain their original pins and do not describe a fresh census of the corrected releases.

project record

Evidence	Policy interfaces and synthetic loader checks
Status	In Progress
Timeline	2026
Role	Archive qualification, command-response analysis and retained simulation tooling
Tools	Python, NumPy, F110 datasets, ROS 2, CalculiX

← All projects