🏋️ AI Fitness Coach: Squat Counter with Pose Estimation

I. Learning Objectives

Understand Applications: Explore how Pose Estimation is used in smart fitness (e.g., smart fitness mirrors, motion-sensing training, athletic assessment).
Master AI Logic: Learn how AI vision sensors identify human poses and judge specific movements.
Action Recognition: Learn to recognize a "Squat" by tracking changes in key pose data.
System Integration: Create a complete AI system: Motion Detection → Counting → Visual Feedback.
Advanced: Understand human body position change → direction judgment → robot turning logic.
Computational Thinking: Develop state machine thinking, logical reasoning, and debugging skills.

II. Preparation

1. Hardware

Controller: LEGO SPIKE Prime / EV3 Hub × 1
AI Vision Sensor × 1
- Mode: Pose Estimation (Pose Mode)
- Positioning: Facing the user, full body or lower body visible
Display: Hub Screen (to show count)
Robot Shell: LEGO-built "Fitness Coach" for aesthetic purposes; no motors required

2. Software

Programming Environment: Spike App / EV3 / RobotCode
AI Model: Pose Estimation (Human Keypoints)

III. Teaching Process

1️⃣ Introduction (5 Minutes)

Guiding Questions
- How does a coach know your squat is "low enough"?
- How do smart fitness mirrors automatically count reps?
Topic Reveal
- Today, we are building a 🏋️ AI Fitness Coach that counts squats automatically.
Key Concept
- We aren't pressing buttons to count; the AI "sees" the action and counts.
- "Besides squatting up and down, people also move left and right. Should the coach always face you?"

2️⃣ Scenario & Principles (5 Minutes)

Key Features of a Squat
- Body moves: Standing → Squatting → Standing
Pose Changes
- Body height decreases significantly
- Angle of legs/hips changes
📌 Simplified Logic:
- "Body goes down + Body comes back up = 1 Squat"

3️⃣ Building & Setup (15 Minutes)

Installation Requirements
- AI Vision Sensor: Place on robot "head" or tripod, 40–70 cm high, facing trainee
- Distance: 1.5–3 meters from trainee
- Field of View: Head + Torso + Thighs
Port Configuration
- Left Leg (Motor): Port E
- Right Leg (Motor): Port F
- AI Vision Sensor: Port A
- Others: Optional/flexible
📌 Teacher Tips
- Keep background simple
- Avoid multiple people in the frame
Arms & Aesthetics
- Robot arms can be designed freely to give the coach personality

IV. Programming (25 Minutes)

1️⃣ Core System Logic

This system uses a State Machine model:

AI Pose Recognition
Determine Current State (Standing vs Squatting)
Detect State Change
Count +1
Display Results

2️⃣ Squat Judgment Rules

State	Description
STAND	User is standing upright
SQUAT	User has lowered their body

How the AI Judges:
- Sensor provides Y-coordinate of body center or hip
- SQUAT: Body Height < Squat Threshold
- STAND: Body Height > Stand Threshold
- Thresholds should be pre-tested by teacher

3️⃣ Counting Logic

❗ Only count a rep when user moves from SQUAT → STAND.
Prevents the counter from incrementing while the person is just holding the squat.

4️⃣ Pseudo-code Example

text

INITIALIZE:
  squatCount = 0
  lastState = "STAND"

LOOP FOREVER:
  currentHeight = Get Human Y-Coordinate from Sensor

  IF currentHeight < squatThreshold THEN:
    currentState = "SQUAT"
  ELSE:
    currentState = "STAND"

  // Did they just stand back up?
  IF lastState == "SQUAT" AND currentState == "STAND" THEN:
    squatCount = squatCount + 1
    DISPLAY squatCount

  lastState = currentState

📌 Key Teaching Point

lastState acts as the robot's "memory". Without it, the robot would count every millisecond spent in SQUAT.

5️⃣ Code Notes

SPIKE Example

alt text

Left Program: Handles movement; person moves left → coach moves left. Adjust motor direction if needed.
Right Program: Handles counting with anti-double-count logic.

RobotCode Example

alt text

Hundreds Digit: Counting enable (0 = no, 1 = yes)
Units + Tens Digit: Turning angle, center point = 50°

V. Demonstration & Testing (10 Minutes)

Student Rotation: Each student does 3–5 squats
Observe:
- Accuracy of counting on Hub screen
- Optional: Robot turns to face user as they move left/right

Troubleshooting Guide

Issue	Potential Cause
Under-counting	Squat not deep enough
Over-counting	Body jitter/shake triggers multiple counts
No detection	Too close/far from sensor; body cut off

VI. Extension & Thinking (10 Minutes)

System Upgrades:
- Precision: Distinguish half-squat vs standard (lower threshold)
- Safety: Detect too fast movements (minimum time interval)
- Versatility: Support other exercises (Jumping Jacks, Lunges)
- Evaluation: AI Fitness Score (Reps + Speed + Stability)

VII. Teaching Summary

Skills Mastered

Pose Estimation applied in fitness
State Machine + Counting logic

Key Understandings

AI is not magic; it's rules + data

Mindset Development

Engineering thinking for real-world AI applications

VIII. Classroom Presentation Material

Student demonstrations of squat counting
Real-time count display on Hub
Robot turning to follow user movement (optional)

🎉 Complete AI Fitness Coach Lesson

🏋️ AI Fitness Coach: Squat Counter with Pose Estimation ​

I. Learning Objectives ​

II. Preparation ​

1. Hardware ​

2. Software ​

III. Teaching Process ​

1️⃣ Introduction (5 Minutes) ​

2️⃣ Scenario & Principles (5 Minutes) ​

3️⃣ Building & Setup (15 Minutes) ​

IV. Programming (25 Minutes) ​

1️⃣ Core System Logic ​

2️⃣ Squat Judgment Rules ​

3️⃣ Counting Logic ​

4️⃣ Pseudo-code Example ​

📌 Key Teaching Point ​

5️⃣ Code Notes ​

SPIKE Example ​

RobotCode Example ​

V. Demonstration & Testing (10 Minutes) ​

Troubleshooting Guide ​

VI. Extension & Thinking (10 Minutes) ​

VII. Teaching Summary ​

Skills Mastered ​

Key Understandings ​

Mindset Development ​

VIII. Classroom Presentation Material ​