RoboRacer

Raceline-following controllers for the FDCL RoboRacer 1/10-scale autonomous car. The car is localised by a Vicon motion-capture system and commanded via SiK radio over serial. Three independent controllers share the same hardware interface, raceline format, dashboard, simulator, and lap counter — only the steering law differs.

---

Pick a controller

Controller	Type	When to use	Detailed setup
PID	Reactive (no horizon, no model)	Baseline. Quickest to run. CLI-tunable gains — no recompile.	README_PID.md
MPPI	Stochastic sampling MPC (300 rollouts)	Production controller. Currently the best-behaved on hardware.	README_MPPI.md
MPC	Deterministic NMPC (acados, SQP-RTI)	Smoothest predicted steering trace; needs a one-time acados build.	README_MPC.md

All three implement the same kinematic-bicycle model and read the same raceline CSV (raceline.csv for the oval, figure_eight.csv for the figure-8 test track).

---

Quick start (simulator)

No hardware required. Verify the controller in the bundled simulator first:

# Cheapest — no extra deps
python3 pid_racing.py  --simulation --raceline figure_eight.csv --laps 3

# MPPI — numpy only, no compilation
python3 mppi_racing.py --simulation --raceline figure_eight.csv --laps 3

# MPC — first run triggers a ~30 s acados compile
python3 mpc_racing.py  --simulation --raceline figure_eight.csv --laps 3

All three open the same live matplotlib dashboard: CTE, heading error, velocity error, track view, controller-specific diagnostic panel, and steering indicator.

---

Repository layout

roboracer/
├── README.md                ← this file
├── README_PID.md            ← PID controller usage and tuning
├── README_MPPI.md           ← MPPI controller usage and tuning
├── README_MPC.md            ← MPC controller usage and tuning, acados setup
│
├── pid_racing.py            ← PID baseline (Stanley-style)
├── mppi_racing.py           ← MPPI controller (numpy)
├── mpc_racing.py            ← Nonlinear MPC controller (acados + CasADi)
│
├── raceline.csv             ← Oval track raceline
├── figure_eight.csv         ← Figure-8 track raceline
├── gen_figure_eight.py      ← Regenerates figure_eight.csv
│
├── serial_test.py           ← Serial-link sanity check
├── sik_test.py              ← SiK radio sanity check
├── waypoint_nav.py          ← Standalone waypoint follower (legacy)
│
├── notes/notes.tex          ← Mathematical overview (model, all three controllers, results)
├── vicon/                   ← Vicon tracker library (C++/pybind11 + VRPN)
└── acados/                  ← acados submodule (only needed for mpc_racing.py)

---

Documentation

• Per-controller usage, setup, and tuning: see the three README_*.md files above.
• Mathematical derivation of the vehicle model, both predictive controllers, and the PID law: notes/notes.tex (compile with pdflatex notes.tex).
• Hardware bring-up procedure (Vicon, SiK radio, serial protocol): covered in README_MPPI.md and README_PID.md — same for all three controllers.

---

Hardware overview

• Vehicle: 1/10 scale RC car, kinematic bicycle parameters L = 0.324 m, δ_max = 0.44 rad, v_max = 15 m/s.
• Localisation: Vicon motion-capture system (subject name UGV, server 192.168.11.2).
• Communication: SiK radio over USB serial at 230400 baud; custom CRC-16 framed protocol.
• Control rate: 40 Hz (time.sleep(0.025) per loop) — all three controllers match this period.

---

Emergency stop

Every controller's finally block sends a zero-throttle, centred-steering packet on normal exit, Ctrl-C, or unhandled exception. Keep a finger on Ctrl-C for the first runs.