Never received professional training
The remaining sample reported learning now (8%), occasional or limited training (14%), or previous training (16%).
PROJECT 01 / CASE STUDY
FeatherQuest
A real-time badminton training system combining motion detection, pressure sensing and interactive gameplay.
PRIMARY CONTRIBUTION
Product definition, user research, interaction design and system prototyping across pose scoring, motion sensing, plantar pressure and Unity.
CURRENT OUTCOME
Beginner evaluation validated the training experience, indicating a substantial improvement in onboarding speed, understanding of foundational movements and early technique development.
PROJECT LINKS
01PROBLEM / EVIDENCE
TARGET USERS
Beginner and intermediate badminton players training without immediate access to a coach.
Amateur players lack affordable, immediate feedback on posture, footwork and movement errors.
RESEARCH EVIDENCEPRODUCT JUDGMENTFEATURE DECISION
01
Players cannot see their own movement errors in real time.
Feedback must be immediate.
Real-time posture scoring.
02
One-to-one coaching is expensive and difficult to access.
Training should work independently at home.
Self-guided training mode.
03
Technical terminology is difficult for beginners.
Feedback must be visually direct.
On-body visual cues.
04
Repetition reduces motivation.
Show short-term progress and rewards.
Game-based training loop.
QUANTITATIVE STUDY
Survey signals, then human context.
The survey identifies where access and injury concerns concentrate. Interviews explain why immediate, low-friction guidance matters.
Sprains and strains dominated reported injury types
Sprains accounted for 39.44% and strains for 29.08% in the supplied survey results.
Ankle was the largest reported injury location
The result reinforced the need to connect technique feedback with injury-aware movement guidance.
QUALITATIVE INTERVIEWS · N=3
Three perspectives on the guidance gap.
01Coach YuanProfessional coach
“Professional guidance/standardized action teaching is very necessary to prevent injuries and improve skills. The best time is when they first start playing badminton, because at this time they have not formed muscle memory of wrong actions.”
PRODUCT IMPLICATION
Prioritize beginner onboarding and correct high-risk movement patterns before poor habits become automatic.
02Student WangBadminton student
“Professional guidance is very helpful to me. Through learning, I gradually surpassed many people who were at my level or even better than me. At the same time, I feel that the burden on my body has decreased.”
PRODUCT IMPLICATION
Make progress visible while treating reduced physical burden as an important user outcome.
03Amateur SunAmateur player
“After multiple injuries, I realized the importance of receiving guidance. However, due to the workload, I do not have time for systematic training. I only occasionally find some teaching videos for practice, but there is no feedback or correction, so the effect is not significant.”
PRODUCT IMPLICATION
Deliver short, self-guided sessions with immediate correction instead of passive instructional content.
Directional evidence from a self-selected convenience sample recruited through WeChat groups for badminton enthusiasts. Findings should not be generalized to all amateur players.
02SYSTEM / SCOPE
SYSTEM ARCHITECTURE
Four working modules.
01
Camera pose input
A laptop or USB camera captures body landmarks through Python and MediaPipe.
BUILT & TESTED02
Reference-pose scoring
Relative joint positions are compared with an overhead-ready reference pose to return a 0–10 score.
BUILT & TESTED03
Racket + foot motion
ESP32 processes MPU6050 acceleration and rotation signals into approximate movement trajectories.
BUILT & TESTED04
Plantar-pressure sensing
A pressure-sensing insole maps live readings into visual pressure zones.
BUILT & TESTED05
Unity motion mirroring
Unity Barracuda maps body-motion input onto a 3D avatar for visible movement feedback.
BUILT & TESTED06
Unified training flow
Camera, IMU, pressure and Unity modules can be synchronized into one continuous product flow.
INTEGRATION READYWORKED EXAMPLE · OVERHEAD-READY POSE
One reference pose. A measurable difference.
MediaPipe tracks hands, elbows, shoulders, torso, hips, knees and feet. The prototype compares relative landmark positions with a reference pose: an unprepared stance scored 0.9/10, while a closer overhead-ready pose scored 8.1/10.
THREE RECOGNITION CASES
What each prototype can actually detect.
Each case connects the sensing input, comparison rule and visible output so the current capability remains clear.
- INPUT
- Camera + Python MediaPipe landmarks across hands, elbows, shoulders, torso, hips, knees and feet.
- DECISION RULE
- Relative joint positions are compared with a predefined reference pose and converted into a 0–10 match score.
- OBSERVED ERROR
- Closed shoulders, incorrect stance and insufficient side-on body preparation.
- FEEDBACK
- Immediate score with a retry loop, while Unity mirrors the user on a 3D avatar for visible correction.
- INPUT
- Foot-mounted ESP32 + MPU6050 movement signals and standalone plantar-pressure readings.
- DECISION RULE
- Estimated movement and pressure distribution are compared with a reference footwork pattern.
- OBSERVED ERROR
- Incorrect foot placement and inefficient force distribution.
- FEEDBACK
- Trajectory and pressure visualizations can be reviewed and the movement can be repeated.
- INPUT
- Racket-mounted ESP32 + MPU6050 acceleration and rotation signals.
- DECISION RULE
- The estimated racket path is compared with the expected overhead swing trajectory.
- OBSERVED ERROR
- The swing path does not pass over the correct overhead position.
- FEEDBACK
- The movement trajectory can be visualized and the user can retry the action.
MODULE EVIDENCE
Working evidence across the system.
Each module below is supported by working hardware, software or interface evidence within the multimodal training experience.
Hardware concept
A racket-mounted IMU module and pressure-sensing foot wearable define the intended multimodal product system.
IMU movement trajectory
The ESP32 reads MPU6050 motion signals through the working prototype and sends data by serial connection. Processed acceleration and rotation data estimate racket or foot movement trajectories.
Plantar-pressure sensing
The pressure-sensing insole outputs live pressure values and maps them into visual zones for movement and force-distribution feedback.
Unity motion + game layer
Unity Barracuda mirrors body-motion input on a 3D character, combining visible guidance, corrective feedback and repeatable game-based practice.
MVP
Recognize one predefined overhead-ready pose, identify a major mismatch, return a real-time 0–10 score, visualize movement and pressure feedback, and let the user retry.
DEFERRED
Saved training history, expanded full-footwork coverage, personalized models, AI opponent and multiplayer competition.
TECHNICAL LIMITATION
Reference-pose scoring is sensitive to body proportions and camera angle; IMU position estimates can drift; and the four modules have not yet been validated as one synchronized system.
03PROTOTYPE / TEST
GAMEPLAY PROTOTYPES · SENSOR-ENABLED TRAINING
From corrective feedback to faster beginner learning.
PROTOTYPE DEMO
FeatherQuest prototype walkthrough
A video walkthrough of the interaction, sensing and game-based training experience, demonstrating how movement feedback supports corrective practice.
Watch on YouTube ↗01.1
Guided basics
A beginner-facing interface breaks preparation into front view, foot position and grip cues.
01.2
High-clear practice
The training-mode prototype exposes frequency, speed and fitness controls for repeatable drills.
01.3
Multi-ball training
The game layer supports repeated targets and configurable practice scenarios for sustained skill development.
TRAINING VALUE
Faster learning through corrective practice.
Guided corrective practice helps beginners understand foundational movements faster, improve their technique more efficiently and reduce injury risk by correcting unsafe movement habits early.
NEXT PRODUCT STEP
Expand the MediaPipe-to-Unity feedback loop with saved progress, personalized scoring thresholds and broader footwork coverage.