PUNCH N FIT – AI-POWERED BOXING PERFORMANCE SENSOR
AI-Driven Boxing Performance Sensor
CATEGORY
- Wearable
- Sports Tech
- IoT
Seun Ola
366 Degree Fitness Studio

SERVICES
- Feasibility & Product Design
- Full-Cycle Product Development
- Hardware & PCB Engineering
- Mechanical Design
- Firmware Development
- Prototyping (EVT, DVT, PVT)
PROJECT TIMELINE
THE CHALLENGE
360 Degree Fitness needed a compact, wrist-mount sensor that can automatically recognize and count boxing punches—jabs, crosses, hooks, uppercuts—without a coach or complex camera system. Requirements included: Edge AI Inference on an nRF52 MCU to classify punch type at ≤10 ms latency; 9-Axis IMU (accelerometer, gyroscope, magnetometer) sampling at 1 kHz for fine-grained motion data; Low Power to deliver 24-hour operation on a coin-cell battery or rechargeable Li-Po; Reliable Connectivity via BLE to stream counts and analytics to a companion mobile app; Comfort & Durability: sweat-proof, shock-tolerant enclosure that stays secure during high-impact training.
- Edge AI Inference: nRF52 MCU to classify punch type at ≤10 ms latency
- 9-Axis IMU: Accelerometer, gyroscope, magnetometer sampling at 1 kHz
- Low Power: 24-hour operation on coin-cell or Li-Po battery
- Reliable Connectivity: BLE to stream counts and analytics to mobile app
- Comfort & Durability: Sweat-proof, shock-tolerant, secure during high-impact training
KEY HURDLES
- •Training an AI model small enough to run in 256 kB flash yet accurate across varied punch speeds
- •Avoiding false positives from non-punch movements (e.g. hand raises, towel wipes)
- •Ensuring robust BLE link through rapid hand motions and in gym environments
- •Packaging electronics into a sub-15 g module that clips comfortably to wrist wraps
THE OPPORTUNITY
Boxers and fitness enthusiasts lack an unbiased, automatic way to track punch volume, speed, and type during training. Punch N Fit offers: Instant Feedback on punch count and classification without manual logging; Performance Insights such as punch speed distribution and fatigue analysis; Motivation & Engagement through real-time metrics, goals, and coach dashboards; Scalable Coaching by enabling remote trainers to monitor multiple athletes simultaneously.
- Instant Feedback on punch count and classification without manual logging.
- Performance Insights such as punch speed distribution and fatigue analysis.
- Motivation & Engagement through real-time metrics, goals, and coach dashboards.
- Scalable Coaching by enabling remote trainers to monitor multiple athletes simultaneously.
FEATURE SETS
- •nRF52832 MCU – Cortex-M4 with 512 kB flash for AI inference
- •9-Axis IMU – 3-axis accel ±16 g, gyro ±2000 °/s, magnetometer ±8 gauss
- •Embedded AI Model – Classifies jab, cross, hook, uppercut, and 'no-punch' gestures
- •BLE 5.0 Connectivity – GATT service for live punch events and session summaries
- •Battery – 120 mAh Li-Po delivering 24 hr continuous use; USB-C recharge in 30 min
- •Haptic Feedback – Short vibration on punch detection to confirm count
- •Local Storage – 64 MB flash buffers session data if phone is out of range
- •IP67-Rated Enclosure – Sealed against sweat and heavy impacts

GALLERY

THE SOLUTION
Axio Electronics delivered a compact, AI-powered boxing sensor with robust BLE connectivity, real-time analytics, and athlete-centric design.

MCU
nRF52832 MCU – ARM Cortex-M4 at 64 MHz with 512 kB flash for on-board AI

AI
Punch Classification – Tiny CNN model distinguishes four punch types plus idle state

BLE
BLE 5.0 Link – Low-latency GATT service streams events to iOS/Android app

IMU
9-Axis Motion Sensor – High-speed accel/gyro/magnetometer sampling at 1 kHz

BAT
Rechargeable Battery – 120 mAh Li-Po for 24 hr use; rapid USB-C charging

HAPTIC
Vibration Feedback – Brief pulse on each detected punch for tactile confirmation

LOG
Local Data Logging – 64 MB flash stores sessions when phone is disconnected

ENV
IP67 Protection – Sealed enclosure withstands sweat, drops, and dust ingress
OUR APPROACH

Mechanical Engineering Strategy
Slim Form Factor under 15 g with curved base to match wrist anatomy
Soft-Touch Overmold for secure fit on hand wraps and gloves
Vent Slots positioned to route away moisture without exposing electronics
Quick-Release Clip for easy mounting and removal

Mechanical Engineering Process
CAD ergonomic studies with pro boxers to optimize strap clearance
3D-printed shells for fit trials under rapid-fire punching
Impact testing to ensure sensor stability through 5 m drop scenarios
Final injection-mold tooling for ABS + TPU hybrid overmold

PCB & Hardware Design
Flexible PCB tracks around wrist curvature
Sensor Fusion: Kalman filter merges accel/gyro/mag for stable orientation
Power Management: Buck regulator for MCU, boost for haptic motor
Antenna Layout: Meandered trace matching 2.4 GHz impedance under overmold
Battery Charger: Linear Li-Po charger with thermal and over-voltage protection
Firmware Design
AI Inference Engine: Quantized CNN runs in 5 ms per frame with CMSIS-NN
Gesture FSM: Debounced state machine to group raw detections into punches
Haptic Controller: Schedules 50 ms vibration bursts on valid punch events
BLE Profile: Characteristic notifications for punch type and timestamp
Data Sync: Batch upload of buffered logs when connection re-established
OTA Updates: In-field firmware flashing via BLE DFU

Prototyping & Validation
EVT: Lab bench tests for classification accuracy using motion-capture ground truth—> 96 % accuracy across all punch types
DVT: In-gym trials with 12 boxers over two weeks—tuned AI thresholds to minimize false positives during defense drills
PVT: 200-unit pilot in four gyms—collected usage data to refine battery life and BLE reliability

Key Challenges & Resolutions
- False Detections from Wrist Flicks— Enhanced CNN training dataset and added motion-severity threshold
- BLE Dropouts During Heavy Impact— Added connection-supervision watchdog and auto-reconnect logic
- Battery Drain Under Continuous Training— Introduced dynamic sampling rate: 1 kHz during active motion, 50 Hz when idle
- Overmold RF Attenuation— Optimized trace geometry and added clearance zones under antenna

PILOT & PRODUCTION
PILOT & PRODUCTION
Following lab validation and intense gym-based field trials, Punch N Fit progressed into pilot deployment across real training environments to assess accuracy, durability, and coaching value:
Gym Pilot: Deployed in 8 gyms with 120 athletes; achieved 92% session compliance and reduced coach feedback loops by 30% through real-time punch tracking.
Design Refinement: Improved wrist-wrap clip and overmold grip after feedback from 2 weeks of high-intensity sparring; enhanced comfort and stability under impact.
BOM Reduction: Achieved 14% cost savings by sourcing an alternative IMU vendor and simplifying the BLE RF chain with integrated filtering.
Certifications: Passed FCC, CE, and Industry Canada (IC) wireless compliance; ASTM F483-certified sweat resistance for wearables.
Manufacturing Test Fixture: Built a custom test rig to validate IMU alignment, AI inference outputs, BLE signal strength, and battery charge-discharge cycles before shipping.

How It Works
Clip Punch N Fit onto wrist wrap or glove strap
Power On via single button; blue LED indicates ready and BLE advertising
Start Session in the mobile app; live punch count and type appear in real time
Receive Haptic Pulse on each detected punch for instant confirmation
Review Analytics post-session: counts, speed distribution, and fatigue curve
WHAT CAN
WE CREATE
TOGETHER
DOWNLOAD THE PUNCH N FIT CASE STUDY (PDF)
Client Testimonial


Seun Ola
366 Degree Fitness Studio
They provided what was necessary to get through the project’s hoops. Prototype completed successfully. The team overcame issues and communicated well (virtual meetings and email). Client praised communication and skill set.
Axio Electronics