Back to Projects

Paper · Control Systems

PaceGuide

Intelligent race pacing via GPS-based control systems — designing the feedback loop that tells a runner to speed up, ease off, or hold.

Institution
George Mason University
Department
Systems Engineering & Operations Research
Tooling
MATLAB / Simulink
Length
9 pages, 30 figures

Abstract

Elite marathoners get a pace car. Eliud Kipchoge’s sub-two-hour attempt was held to schedule by a vehicle driving a precise, unwavering speed. Everyone else glances at a watch and does mental arithmetic mid-race — which is why recreational pacing falls apart.

This paper builds that pace car into a GPS smart watch. The system reads position, derives velocity, compares it against the target, and returns one of three instructions: faster, slower, or maintain. Four progressive Simulink models take it from an idealized proportional controller to one that survives discrete sampling and noisy GPS fixes.

The finding that matters: velocity feedback alone always leaves a residual distance error, because small speed deviations integrate into meters of shortfall over an hour. Adding a position-error integrator closes that gap — the tuned dual-input controller lands on 10 km at exactly 3600 seconds.

First page of the PaceGuide paper

View full paper →

Four Models

Each model keeps the previous one’s structure and removes one convenient assumption, so the cost of every real-world constraint can be read off in meters.

01

Proportional Control

Instantaneous velocity feedback with gain K. Sweeping K = 0.1, 1, and 6 exposes the trade between responsiveness and stability.

K = 1 finishes 3 m short; K = 0.1 falls 28 m short.

02

Discrete-Time Sampling

Velocity is no longer known instantly — it is differenced from positions sampled every Δt. Tested at Δt = 1 s, 4 s, and 10 s.

Δt = 10 s costs 13 m of the 10 km target.

03

GPS Measurement Error

Zero-mean Gaussian noise (σ = 2 m) is injected into each position fix, so the derived velocity inherits sensor uncertainty.

Noise amplifies as Δt shrinks — smaller windows are worse.

04

Dual-Input Control

A third-order controller adds an integral position term (Ki) on top of velocity gain (Kp) to correct accumulated drift, not just instantaneous speed.

Ki = 1.5 × 10⁻⁵ hits 5 km at 1800 s and 10 km at 3600 s exactly.

Simulation Parameters

10 km

Race distance

60 min

Target finish

2.778 m/s

Target pace

σ = 2 m

GPS error

Authors

  • Hocine FilaliSimulink models, Model 4 analysis, conclusion
  • Amr HamzaModel 1 and Model 2 analysis
  • Adewale Adekambi— meAbstract, introduction and background, Model 3 analysis

Read the full paper

Nine pages covering the block diagrams, governing equations, and simulation output for all four models.

PaceGuide.pdf