Lab 2 — USV PID Tuning

Week 7 · Spring 2026

Status: in progress through Week 7. This week is data collection; analysis and performance characterization happen in Week 8.

Objectives

Tune the autopilot’s closed-loop control to maximize speed and yaw-rate tracking performance. By the end of this week each team has step-response data for three experiments × two channels = six datasets.

Experiment What changes What stays the same
A — Baseline Lab preset tweaks (rate limiters off, FF off, throttle baseline off); ArduRover stock PID gains —
B — FF Feedforward term restored Stock PID gains
C — Tuned Students tune PID gains for best response Feedforward kept on

For each experiment, both the speed (surge) loop and the yaw-rate (steering) loop are exercised with step inputs, and the resulting closed-loop responses are logged and quality-checked.

This week’s deliverable: clean step-response logs for all six cases, with a brief quality-check note per dataset.

Next week: process the data with closedloop_stepinfo.m to quantify rise time, settling time, overshoot, and steady-state error — and use those metrics to compare the three experiments.

Architecture overview

The lab tunes the two innermost feedback loops in ArduRover’s ACRO mode:

  • Speed (surge) loop — PID under ATC_SPEED_*, target from RC throttle scaled by CRUISE_SPEED, measurement from GPS.
  • Yaw-rate loop — PID under ATC_STR_RAT_*, target from RC steering scaled by ACRO_TURN_RATE, measurement from EKF/IMU.

Block diagram, parameter list, and logging channels are in Architecture details.

Procedure overview

Working draft — see Procedure details.

At a high level:

  1. Confirm preset parameters and the experiment-specific tuning state on the boat.
  2. Run a series of throttle and steering step inputs in ACRO mode.
  3. Pull the .BIN log, convert with bin2mat.py, and inspect with proc_closedloop_response.m.
  4. Verify the data passes the quality check before returning to the dock.

End state — what “done” looks like this week

For each of the six datasets, you should be able to plot — using closedloop_stepinfo.m — a step-response figure that shows:

  • Setpoint (commanded step) overlaid on measured response
  • Control effort (normalized throttle or rudder) on a linked time axis
  • PID internals (PIDT_* for speed, PIDS_* for yaw rate) showing target, actual, P/I/D/FF terms, and total output

If those plots look clean — clear step, response settles, no clipping, no logging gaps — the dataset is good. If anything looks off, plan to recollect before leaving the field.

Materials