graymarshall.dev — portfolio
Back to index
○ In Development/2026 — Present

Longhorn Racing Autonomous Platform.

FSAE driverless simulation + control stack for Longhorn Racing's autonomous program.

Longhorn Racing Autonomous Platform
FIG. · lhr-autonomy
Year 2026 — PresentStatus ○ In DevelopmentStackROS 2PythonGazeboRVizPure Pursuit
I.
Summary
Overview

A ROS 2 workspace implementing the full FSAE driverless pipeline — procedural cone track generation, FOV-limited sensor simulation, centerline extraction, pure-pursuit control, the competition state machine, and a metrics harness — running on a pluggable simulation backend (lightweight kinematic model or Gazebo physics). Built as the starting point for next season's autonomous program at Longhorn Racing.

II.
Margin
Notes
  • §Built a 9-package ROS 2 workspace covering the full driverless loop — track generation, sensor simulation, centerline extraction, mission state machine, pure-pursuit control, vehicle simulation, metrics, and a Gazebo physics backend — orchestrated by a single `lhr_demo` launch file.
  • §Procedural cone track generator (`lhr_trackgen`) — Catmull-Rom spline autocross courses parameterized by seed, waypoint count, radius, jitter, track width, and cone spacing, publishing ground-truth blue/yellow cone markers.
  • §FOV-limited sensor simulation (`lhr_sensor_sim`) — configurable field-of-view, max/min range, Gaussian position noise, and false-negative rate; accumulates detections over time and publishes bright/dim seen-vs-unseen visualization plus an FOV frustum.
  • §Pure-pursuit controller (`lhr_control`) with a curvature-driven speed law — `v = clamp(sqrt(a_lat_max / |κ|), v_min, v_max)` — plus separate longitudinal accel/decel limits. Parametric lookahead, wheelbase, and curvature estimation window.
  • §FSAE driverless state machine (`lhr_mission_manager`) — `Off → Ready → Driving → Finished` with an `Emergency → Off` reset path. Mission selectors for inspection / autocross / acceleration / skidpad / EBS-test / manual. The control node is gated by mission state and only publishes drive commands when `DRIVING`.
  • §Metrics harness (`lhr_metrics`) — live cross-track error, off-track threshold events, and lap detection with start-radius + hysteresis + min-lap-time filters. Emits lap-complete events back to the mission manager and appends a per-run row to `data/metrics.csv`.
  • §Two interchangeable simulation backends: a lightweight kinematic bicycle model for fast iteration, and a full Gazebo physics world generated from the same track definition — identical downstream ROS 2 topic surface, so the upper stack doesn't know which is running.
  • §Developer ergonomics: Humble/Jazzy auto-detecting build and run scripts, pre-configured RViz layout, PlotJuggler integration for curvature / speed / state debug plots, single-command full-stack launch.
Other work
Live · 2025 — Present
Longhorn Sim Racing Platform
Events, Membership & Competition Management Platform
Next.js 16 · TypeScript · Prisma
In Development · 2024 — Present
Formula SAE EV Telemetry System
Distributed Real-Time Vehicle Telemetry & Analytics Platform
Next.js · TypeScript · Python
In Development · 2025 — Present
In-Vehicle Driver Dash System
Real-Time Driver Display for Vehicle State, Timing, and Energy Data
Real-Time Systems · WebSockets · HMI