RAD

RESEARCH AND DESIGN

RTK-GNSS Project

MEASURING CORNERING PERFORMANCE IN COMPETITIVE CYCLING

A GNSS/IMU-based systems approach. By Will Seitz.

ABSTRACT

This paper outlines the engineering considerations for a 20-rover RTK-GNSS and IMU telemetry system that measures cornering performance in cyclists on a closed circuit or time trial course. It covers the hardware, the software needed to survey a course and analyse rider performance against it, the people needed to run the project credibly, and the technical limitations to expect.

1. INTRODUCTION

Cornering is one of the least instrumented parts of cycling. Power meters and heart rate data are everywhere in the peloton, but cornering metrics are not. A fleet of RTK-GNSS/IMU rovers tied to one command computer makes this data practical to collect for a squad of 20 riders instead of one athlete at a time.

The goal is more than logging. It is to derive an optimum line and speed profile for each corner from real ridden data, and to benchmark individual riders against it. With that benchmark, riders can decide how to cut the time they lose in corners.

2. HARDWARE

2.1 Rover unit (x20)

Each rover is built around the u-blox ZED-X20P, an all-band, all-constellation GNSS receiver. It supports RTK at up to 25 Hz with centimeter-class accuracy (about 1 cm + 1 ppm CEP) and fuses IMU data in firmware, so it outputs roll, pitch, and attitude directly instead of only raw acceleration. Each rover is specified at under 500 mL in volume and under 2 kg in mass. Key considerations:

Antenna placement. Mount the antenna rigidly on the frame (seatpost or stem), not on the rider's body, so the measured position follows the bike rather than an unstable point. The antenna is not at the tire contact patch, so measured lean angle and line are proxies, not ground truth (see section 5).
Time synchronization. The module's time-pulse output should hardware-timestamp every fused epoch, so all 20 rovers agree to well inside one sample period (40 ms at 25 Hz). Without this, comparing riders at the same instant is unreliable.
Power and mounting. Sessions of several hours need a battery margin beyond the ride time. A vibration-damped mount matters, because antenna movement relative to the bike adds noise directly to the radius and lean angle calculations.

2.2 Base station

A single surveyed, fixed-position RTK base station broadcasts RTCM correction data to all the rovers. Alternatively it logs raw data for post-processed kinematic (PPK) correction when real-time is not required. The quality of the whole dataset is limited by the quality of this one survey. An error here carries into every rover and every session.

2.3 Central command computer

The command computer hosts the base station, distributes corrections, and ingests and logs the data. Whether it needs a live telemetry radio link (for example LoRa) depends on whether you need to monitor sessions as they happen. For post-session analysis only, a dock-and-download workflow is enough and much cheaper.

3. SOFTWARE

The system is split into two programs that run at different cadences.

3.1 Program 1: Course boundary mapping

This runs once per course, or after any resurfacing or reconfiguration. A rover is walked or ridden slowly along the inner and outer boundaries. The tracks are smoothed with a spline fit and used to derive a centerline, a track-local arc-length and lateral-offset (s, n) coordinate frame, a corridor-width function w(s), and corner-zone definitions for entry, apex, and exit. The output is one versioned course definition file that every later session refers to.

3.2 Program 2: Corner analysis

This runs every session. It reads each rover's fused position, velocity, and attitude log, projects it into the course's (s, n) frame, and splits it into laps and corners. For each corner it computes:

Entry, apex, and exit speed.
Fitted radius.
IMU-measured lean angle.
Braking and acceleration onset points.
Lateral line compared with the reference centerline.

It combines these across riders and laps, and finds an optimum line for each corner with two complementary methods:

Empirical (Pareto). Pool every rider's (offset, speed) pairs at each point along the corner, pick out the locally best combinations, and stitch them into a smooth candidate path.
Physics-based. A minimum-time path and speed solver, limited by the surveyed corridor width and a friction ceiling. The ceiling comes from the highest sustained lateral acceleration seen without a slip in the real IMU data, not from a textbook tire coefficient.

Comparing the two shows the gap between what riders have actually achieved and the theoretical ceiling.

4. EXPERTS NEEDED

No single discipline covers this project. Realistically it takes:

GNSS and surveying specialist. To run the base station survey and course mapping correctly. Errors here are invisible in the data but corrupt everything after them.
Embedded and firmware engineer. For rover configuration, power management, time sync checks, and telemetry reliability.
Software and data engineer. To build the ingestion pipeline, coordinate transforms, corner segmentation, and database.
Applied mathematician or controls engineer. For arc fitting, filtering, and minimum-time path optimisation. This is a hard numerical problem, closely related to racing line optimisation in motorsport.
Vehicle dynamics and motorsport. The minimum-time path under a friction circle is well studied in automotive work and transfers directly. But a bicycle is a single-track vehicle with rider lean coupling and no independent steering and lean control. A straight port of car racing line code will misrepresent the physics unless it is adapted.
Sports scientist or cycling coach. To interpret what the numbers mean for training and technique, and to check that "optimal" outputs are sensible advice for a human rider.
Statistician. So that comparisons across riders are meaningful, with proper variance, confidence, and no false precision.

5. LIMITATIONS

Antenna to contact patch offset. The antenna sits on the frame, well above and ahead of or behind the tire contact patches. A leaned bike's antenna traces a different arc than its tires. The IMU helps, because it measures actual frame roll, but lean angle and line are still frame-referenced. That is acceptable for most coaching, but it does not replace a true tire force measurement.
GNSS environment. RTK performance drops under tree canopy, near buildings, in cuttings, and under bridges. Multipath near barriers is a known failure exactly where cornering data matters most. Some corners in each session will have a degraded fix. Those need to be flagged and discarded or down-weighted, not averaged in.
Sustained 25 Hz RTK. Manufacturer rates are for good conditions. Holding an RTK-fixed 25 Hz across 20 units in a real outdoor RF environment, with antennas shadowing each other in a bunched group, has not been validated and should be tested before relying on it for competition-grade conclusions.
Human and environmental variability. Wind, tire choice, tire pressure, road temperature, and fatigue all change achievable cornering speed, and position and attitude data do not fully capture them. Comparing one rider's Tuesday morning corner with another's Thursday afternoon corner compares conditions as much as skill.
The optimum is bounded by what has been ridden. The Pareto method cannot find a line nobody has tried. The physics method is only as good as the friction ceiling taken from observed data, and if no rider has approached the true grip limit it will underestimate it.
Isolated riding, not racing. Drafting, group dynamics, tactics, and reacting to other riders are absent from single-rider analysis. A line that is best alone may be irrelevant or unsafe in a bunch.
Cost and logistics. Twenty rovers means twenty units to charge, mount, calibrate, and download every session. This overhead, more than the GNSS technology, is often the real bottleneck.
Validation. Trusting derived lean angle and speed means checking them against an independent reference, such as a calibrated reference rider, video, or track-mounted timing, for at least some sessions.

6. CONCLUSION

The proposed system is technically achievable with current commercial RTK-GNSS and IMU hardware, and it would be a real advance on the near-total absence of corner-level cycling telemetry today. Its outputs are best treated as high-resolution, well-grounded estimates, not exact physical measurements. The antenna offset, environmental, and human-variability limits are structural, and better software alone will not remove them. A credible deployment budgets time and expertise for course survey quality, a validation pass against an independent reference, and honest handling of degraded-fix data, instead of treating raw rover output as ground truth.

back to home

rdand.net is a website of RAD Performance.

+1 [205] 902-5236
will@rdand.net