What 279 recorded rides say about how Earth Rovers really move

Description and use

The pilot now correctly interprets angular commands, avoiding steering errors and improving checkpoint success, though GPS lag correction was not adopted due to negligible benefit.

Problem

The Earth Rover SDK lacks clear specifications for angular direction, speed, unit conversions, and sensor latency, causing pilots to misinterpret commands and fail to reach checkpoints.

Outcome

The pilot now correctly interprets angular commands, avoiding steering errors and improving checkpoint success, though GPS lag correction was not adopted due to negligible benefit.

Workflow

• 1. Downloaded 24 zip parts via HTTP range requests to read control, GPS, and IMU CSVs.
• 2. Extracted 279 usable rides and ~1 million control rows from ~317 MB of data.
• 3. Analyzed angular direction using a sign test against GPS course.
• 4. Determined angular speed, linear speed, wheel response time, GPS lag, and wheel radius from the data.
• 5. Updated the pilot to interpret positive angular as left turns and adjusted spin duration.
• 6. Validated changes by simulating 360 missions and comparing checkpoint success rates.

Evidence

The SDK doesn't say which way a positive <code>angular</code> turns, how fast the rover turns or drives per command, what units the gyroscope reads, or how late the GPS is.

Source

A positive <code>angular</code> turns the rover <strong>left</strong>: 440 to 57 in a sign test against the GPS course.

Source

It turns at about <strong>30°/s per unit of <code>angular</code></strong>. The gyroscope's x axis is yaw and reads degrees per second, counter-clockwise positive.

Source

It covers about <strong>0.8 m/s at full <code>linear</code></strong>.

Source

The wheels respond about <strong>0.8 s</strong> after a command, and the GPS shows the motion about <strong>1.5 s</strong> after the wheels.

Source

The effective wheel radius is <strong>0.064 m</strong>, and the logged RPMs are signed.

Source

An uncalibrated real rover is now driven with a positive <code>angular</code> turning left.

Source

The measurement exposed a bug. At 30°/s, our calibration's 30 s spin at 0.4 covers only about 350°, short of the whole turn it needs, so it would most likely have failed on a real rover.

Source

Method, numbers and code: <a href="https://github.com/munimtechnologies/earth-rover-pilot/blob/main/FINDINGS-frodobots-2k.md" rel="noopener noreferrer">FINDINGS-frodobots-2k.md</a>.

Source