New Framework Bridges Sim-to-Real Gap in Autonomous Racing Under Real-Time Constraints

Autonomous racing pushes vehicles to their absolute limits—high speed, tight stability margins, and hard real-time deadlines. Controllers that ace simulation often degrade abruptly on physical tracks due to interacting effects of dynamics mismatch, estimation delay, and execution-layer latency. A new paper from researchers Hossein Maghsoumi and Yaser P. Fallah (accepted at IEEE VTC2026-Fall) frames this sim-to-real gap as a full-stack, real-time systems problem. They introduce a structured three-layer perspective—Physical, Cyber, and Execution—to analyze how mismatches propagate and amplify through closed-loop feedback. The work goes beyond simple lap time, proposing diagnostic metrics such as “performance flip” (where a controller that wins in simulation loses on the real track), stability-oriented measures, sensitivity to delay and noise, and latency distribution characterization.

The team synthesizes mitigation strategies from a deployment-oriented viewpoint, emphasizing execution-aware and delay-aware design. For example, controllers must account for the actual latency of perception, planning, and actuation loops rather than assuming ideal timing. The paper also outlines benchmarking guidelines that enable reproducible and fair sim-to-real evaluation under compute and timing constraints—a critical step for the field. By clarifying cross-layer failure mechanisms and providing practical design principles, the framework helps developers build autonomous racing systems that can operate safely near dynamic limits. For professionals in robotics and autonomous systems, this research offers concrete tools to diagnose why perfectly good sim controllers break in the real world and how to fix them.