How Does PoinT GO Develop Algorithms for Accuracy?

Why Accuracy Matters

In sports science, the value of any measurement device ultimately comes down to accuracy. No matter how convenient or affordable a tool is, if the measurements cannot be trusted, they cannot serve as the basis for training decisions.

PoinT GO performs a wide range of measurements — VBT, jump, RSI, ROM, isometric testing, weightlifting analysis, and more — all with a single IMU (Inertial Measurement Unit) sensor roughly the size of a fingertip. So the question naturally arises: can you really trust the numbers from such a small sensor?

In this article, we openly share the process our team follows to ensure algorithm accuracy.

Core Principle: Validation Against Gold-Standard Equipment

The core principle behind PoinT GO's algorithm development is simple:

Use data from equipment that is physically guaranteed to be more accurate as the ground truth.

An IMU sensor measures acceleration and angular velocity, then derives velocity, displacement, and height through calculations. Integration errors, drift, and noise can all occur during this process. That's why we use direct measurement devices — equipment that obtains the target physical quantity without intermediate calculations — as our reference standard.

Jump & RSI: Comparison with Force Plates

Why Force Plates Are the Gold Standard

A force plate directly measures the force exerted by the feet on the ground. Using Newton's second law (F = ma), subtracting body weight from the measured force yields net acceleration, which can be integrated to accurately calculate velocity and displacement. This is why force plates are recognized as the international standard measurement equipment for jump height, flight time, RSI, and related metrics in sports science research.

Validation Process

The PoinT GO team refines its jump/RSI algorithms through the following process:

1. Simultaneous measurement: An athlete performs jumps on a force plate while wearing a PoinT GO sensor.

1. Data collection: Force-time data from the force plate and acceleration-time data from the PoinT GO sensor are recorded simultaneously.

1. Result comparison: Key metrics — jump height, flight time, ground contact time, RSI — are calculated from both devices and compared.

1. Error analysis: Systematic bias and random error are separated and analyzed.

1. Algorithm refinement: Based on the analysis, sensor algorithm parameters are adjusted, and when necessary, the algorithm structure itself is improved.

1. Iterative validation: Comparison measurements are repeated with the improved algorithm to confirm accuracy gains.

This is not a one-time process — it's a continuous cycle. Validation is repeated whenever new jump types are added or sensor firmware is updated.

Key Validation Metrics

Metric	Force Plate Method	PoinT GO Method
Jump height	Takeoff velocity → mechanical energy conversion	IMU acceleration integration → velocity → displacement
Flight time	Direct measurement of force = 0 interval	Takeoff/landing detection via acceleration patterns
Ground contact time	Direct measurement of force > 0 interval	Detection via acceleration rate of change
RSI	Jump height ÷ Ground contact time	Combination of above calculations

VBT: Comparison with Tethered Devices

Why Tethered VBT Devices Are the Reference

Tethered VBT devices use linear encoders or cable-based transducers physically connected to the barbell. They measure displacement directly from the cable length pulled, then differentiate by time to calculate velocity.

Because this method directly measures displacement, it is inherently more accurate than IMU-based integration. However, the cable connection makes it cumbersome to use, and certain exercises (e.g., Olympic lifts) are difficult to measure with tethered systems.

Validation Process

1. Simultaneous measurement: A tethered VBT device is connected to the barbell while a PoinT GO sensor is also attached.

1. Varied conditions: Measurements are taken across multiple exercises (bench press, squat, deadlift, etc.), various loads, and different velocity ranges.

1. Rep-by-rep comparison: Mean velocity, peak velocity, and ROM for each rep are compared between devices.

1. Fitting optimization: Statistical methods like Bland-Altman analysis and correlation analysis are used to identify error patterns and optimize the algorithm.

Key Validation Metrics

Metric	Tethered Device Method	PoinT GO Method
Mean velocity	Direct displacement measurement → differentiation	IMU acceleration integration
Peak velocity	Maximum of displacement derivative	Maximum of acceleration integral
ROM	Cable length pulled	Velocity integration

Real-World Algorithm Improvement Examples

Case 1: VBT ROM Calculation Method Transition

Initially, PoinT GO used barometer (air pressure) sensor Position data to calculate barbell displacement. However, comparison with tethered devices revealed that the barometric method was sensitive to environmental changes (air conditioning, door opening/closing, etc.).

We transitioned to an IMU Velocity Integration method — integrating acceleration to obtain velocity, then integrating velocity to calculate displacement. After this change, agreement with tethered devices improved significantly.

Case 2: Jump Detection State Machine Refinement

Comparing with force plate data, we identified detection misses under specific conditions (low jumps, rapid consecutive jumps, etc.). To address this, we fine-tuned the state machine transition conditions in the jump detection algorithm and optimized thresholds across various jump height ranges.

Case 3: Drift Correction Algorithm

To tackle the inherent integration drift problem of IMU sensors, we implemented a Zero-Velocity Update (ZUPT) algorithm. This resets velocity to zero when the barbell is stationary, eliminating accumulated errors. This technique is also used extensively in throws measurements.

Continuous Improvement Cycle

PoinT GO's algorithm development is never "done." We continuously improve accuracy through the following cycle:

1. Field feedback — Real-world usage data and feedback from coaches and athletes

1. Laboratory comparison — Simultaneous measurement with gold-standard equipment

1. Data analysis — Identifying error patterns and root causes

1. Algorithm improvement — Parameter tuning or structural changes

1. Field testing — Validation in real training environments

1. Deployment — Sensor firmware and app updates

A Philosophy of Transparent Development

The PoinT GO team's goal is "small sensor, big accuracy." We acknowledge the physical limitations of IMU sensors while striving to overcome those limitations through the power of algorithms.

To achieve this:

• Comparison with gold-standard equipment is part of our routine development process

• We accumulate quantitative validation data for every measurement type

• Algorithm improvements are made through data-driven decision-making only

• New measurement features always undergo comparison validation first

Performing diverse fitness measurements with a single small IMU sensor is a challenging endeavor. But through constant comparison and validation against the most physically accurate equipment available, PoinT GO is evolving every day to deliver field-reliable accuracy.

Start 8 types of fitness measurement with a single Point Go sensor. We continuously refine our algorithms, targeting gold-standard accuracy.