Home Games Research Shader Sandbox

Research & Development

Moonjump · MotionMatrix research program

Making high-fidelity motion capture wearable.

Moonjump investigates whether sparse body-worn inertial sensors, synchronized over a low-latency wireless network and interpreted by biomechanical models, can produce stable, production-useful 3D motion—without requiring a camera volume.

DocumentR&D position paper
Revision1.1 · July 2026
ProgramMotionMatrix
StatusActive multi-year R&D
00 Abstract

Motion capture beyond the volume.

Optical motion capture can produce high-quality ground truth, but it depends on cameras, line of sight, calibrated spaces and specialist operation. Inertial systems can move beyond that volume, yet introduce different uncertainties: sparse observation, accumulating drift, clock error, radio loss, body coupling and incomplete biomechanical information.

Abstract

MotionMatrix is Moonjump’s applied research program into wearable human-motion measurement. The work spans miniaturized inertial sensing, deterministic body-area networking, calibration, long-duration drift correction, biomechanically constrained motion reconstruction, multi-user capture, real-time analytics and production-pipeline interoperability. The SensorMatrix system is the integrated experimental platform through which these questions are tested together.

Our central question: can a sparse wearable system preserve the timing, identity and biomechanical meaning needed for production-quality 3D motion?

01 Research thesis

The body is observable—but only in fragments.

An IMU measures local acceleration and rotation. It does not directly observe a complete skeleton, joint limits, contact, global position or performer intent. MotionMatrix treats reconstruction as a whole-system inference problem rather than a sensor-export problem.

01

Wearable sensing

Miniaturized flexible sensor modules, body coupling, thermal behavior, power, comfort and environmental durability.

02

Time & transport

Deterministic wireless scheduling, clock synchronization, packet ownership and recovery across many body-worn nodes.

03

Biomechanical intelligence

Sparse-signal pose reconstruction, learned temporal models, body constraints, sensor placement and long-session drift correction.

04

Production validity

Optical ground-truth comparison, multi-user trials, real-time analytics, format fidelity and tests inside animation pipelines.

02 System architecture

From body-worn signal to production skeleton.

Accuracy at the end of the pipeline depends on every stage preserving time, identity, calibration state and uncertainty. A strong model cannot recover information that has already been distorted by fit, synchronization or transport.

MotionMatrix research stackSENSE → SYNCHRONIZE → RECONSTRUCT → VALIDATE
01 · SenseBody-worn inertial modules

IMU measurements, sensor placement, attachment, power, environmental state and local timestamping.

02 · SynchronizeCalibrated body-area network

Clock alignment, deterministic slots, packet identity, radio recovery and sensor-to-segment calibration.

03 · ReconstructBiomechanical motion model

Temporal inference, skeletal constraints, drift correction, multi-user association and live analytics.

04 · ValidateGround truth & production QA

Optical comparison, error analysis, environmental trials, format round trips and studio pipeline review.

03 Core R&D activities

Eight uncertainties, one integrated system.

The core activities are coupled. Module geometry affects radio performance; synchronization affects model quality; drift correction affects long-session usability; and export semantics determine whether technically accurate motion remains useful in production.

A Motion inference

Biomechanical motion model from sensor input

Investigating whether six-to-nine body-worn IMUs can recover stable full-body skeletal motion when temporal models and biomechanical constraints compensate for unobserved joints and sparse sensor placement.

  • Sparse IMU pose reconstruction
  • Optical ground-truth alignment
  • Temporal sequence models
  • Sensor placement optimization
B Session stability

Long-duration drift correction

Characterizing how MEMS drift evolves across extended sessions and testing whether temporal context, cross-sensor consistency and learned correction can reduce recalibration interruptions without adding visible lag.

  • Gyroscope bias estimation
  • Magnetic disturbance detection
  • Learned drift signatures
  • Multi-hour replay evaluation
C Wearable hardware

Miniaturized sensor module v2

Co-designing flexible PCB layout, IMU stability, radio behavior, battery systems, thermal performance and attachment so a smaller module does not trade away signal quality or repeatable body alignment.

  • Flexible PCB and power tree
  • IMU noise and thermal behavior
  • On-body antenna performance
  • Comfort and manufacturability
D Shared capture

Multi-user simultaneous tracking

Testing whether several SensorMatrix suits can operate in one space without radio collision, clock divergence, cross-user packet mixing or identity loss during close performer interaction.

  • Cross-user synchronization
  • Identity persistence
  • Collision-aware scheduling
  • Scene-level interaction timing
E Live interpretation

Real-time motion analytics

Developing a streaming feature and inference pipeline for immediate motion feedback while measuring the trade-off between latency, stability and meaningful biomechanical interpretation.

  • Streaming inference latency
  • Motion-sequence features
  • Stable live feedback
  • Edge processing constraints
F System integration

SensorMatrix advanced development

Integrating hardware, firmware, calibration, networking, models and tools as a versioned system so subsystem gains survive end-to-end operation and failures remain attributable.

  • Configuration compatibility
  • Regression testing
  • Capture observability
  • Field deployment reliability
G Field robustness

Real-world and environmental validation

Measuring the combined effects of heat, cold, humidity, perspiration, mechanical strain, clothing, imperfect fit and electromagnetic interference on tracking quality.

  • Compound stressor testing
  • Environmental compensation
  • Field-to-lab reproducibility
  • Failure-mode classification
H Timing network

Wireless body-area network protocol

Evaluating deterministic TDMA, BLE and UWB strategies for low-latency streaming, tight clock synchronization, packet recovery and multi-user scale across distributed body-worn nodes.

  • Deterministic slot scheduling
  • Clock synchronization
  • Packet-age-aware recovery
  • On-body RF characterization
04 Supporting R&D activities

Research must survive contact with the wearer and the studio.

Five supporting activities establish the operational boundaries needed to interpret the core experiments and move comparable motion through real workflows.

Setup repeatability

Calibration workflow standardization

Guided poses, sensor-to-segment alignment, automated quality scoring and repeatability across body types and operators.

Supporting activity
Production workflow

International studio pipeline testing

MotionBuilder, Maya and Unreal workflows; retargeting behavior; cleanup burden; timing thresholds and studio review criteria.

Supporting activity
Motion semantics

Data-format compatibility layer

A canonical representation and round-trip testing across BVH, FBX, C3D and USD, with skeleton, timing and provenance preserved.

Supporting activity
Operational runtime

Power firmware optimization

Per-state current profiling, dynamic clocks, sensor duty cycles, adaptive radio scheduling and wake-latency validation under live capture.

Supporting activity
Physical reliability

Sensor housing durability

Impact, fatigue, temperature, perspiration, ingress and shielding tests that connect mechanical design directly to signal integrity and fit.

Supporting activity
05 Experimental method

Measure the whole chain, not only the model.

MotionMatrix uses paired sensing and reference systems to separate algorithmic uncertainty from implementation defects. Optical capture provides ground truth where appropriate; instrumented hardware and synchronized logs expose what happened before reconstruction.

Define the uncertainty and testable hypothesis

State the unknown in measurable terms—pose error, drift, synchronization, packet loss, latency, runtime, durability or format fidelity—and define what would support or reject the hypothesis.

Capture synchronized source and ground truth

Record raw IMU streams, calibration state, radio timing and environmental context alongside optical reference motion and performer/session metadata.

Build discriminating prototypes

Compare sensor placements, board revisions, radio schedules, correction models, reconstruction architectures and export mappings rather than optimizing one unchallenged design.

Stress the operating envelope

Change performer, motion type, session duration, user count, temperature, humidity, fit, interference, battery state and studio toolchain.

Evaluate end-to-end

Measure skeletal fidelity, temporal alignment, drift, RF reliability, power, thermal behavior, comfort, cleanup burden and round-trip conversion error.

Preserve traceability

Link datasets, build revisions, firmware/model versions, training logs, evaluation scripts, hardware artefacts and conclusions so results can be reproduced and challenged.

06 Evaluation framework

A useful skeleton is more than a low average error.

A system can look convincing in a short demo and still fail under drift, interaction, radio loss or production conversion. Evaluation therefore spans measurement quality and operational usefulness.

DimensionResearch questionRepresentative evidence
Skeletal fidelityDoes reconstructed motion remain biomechanically plausible and close to optical reference across joints and movement types?Per-joint angular/positional error, contact behavior, pose-flip and failure-case review.
Temporal integrityAre distributed sensors and users aligned tightly enough to preserve rapid motion and physical interaction?Clock offset, end-to-end latency, jitter, packet age and cross-user alignment.
Long-session stabilityHow does accuracy change with time, temperature, magnetic disturbance and repeated movement?Multi-hour drift curves, recalibration events, disturbance trials and correction residuals.
Wireless reliabilityCan the body-area network scale without collision, packet mixing or production-visible loss?Packet-loss distributions, slot timing, RF measurements, recovery behavior and user-count tests.
Wearability & powerDoes the hardware remain comfortable, aligned, thermally safe and operational through a real session?Current profiles, runtime trials, temperature, fit repeatability, fatigue, ingress and wear feedback.
Pipeline validityDoes motion retain timing, hierarchy, metadata and editability in professional tools?Round-trip format tests, retargeting review, cleanup time, provenance checks and studio trials.
07 Open research questions

What remains technically uncertain.

These questions describe the active research frontier. They are not claims of completed performance.

RQ–01

How few body-worn IMUs can recover production-useful full-body motion when biomechanical and temporal priors reconstruct what is not directly observed?

RQ–02

Can drift be separated from genuine performer motion early enough to sustain multi-hour sessions without disruptive physical recalibration?

RQ–03

Which wireless scheduling and synchronization strategies preserve sub-frame timing across many on-body nodes and several co-located users?

RQ–04

How small and flexible can the sensor module become before thermal, RF, power and attachment effects dominate measurement quality?

RQ–05

Can performer identity and interaction timing remain stable when multiple users touch, cross paths or temporarily lose packets?

RQ–06

What combination of metrics best predicts whether motion will remain editable and acceptable after retargeting, conversion and studio review?

08 Evidence & collaboration

A living research program with traceable evidence.

Moonjump maintains an evidence chain from activity-level uncertainty through research briefs, engineering rationale, prototype development, component design, synchronized datasets, evaluation scripts and result interpretation.

Public disclosure boundary

This paper is a public synthesis of Moonjump’s internal MotionMatrix R&D documentation. It deliberately omits confidential reference identifiers, partner material, raw datasets, unreleased electrical and mechanical designs, model weights, security-sensitive protocol details and unverified interim measurements. Numerical thresholds mentioned as research targets are evaluation criteria—not product claims.

Working on wearable sensing, motion intelligence or production capture?

We welcome technically grounded conversations with researchers, studios, hardware specialists, biomechanists and production teams interested in sparse-sensor reconstruction, body-area networking, calibration, validation and motion-pipeline interoperability.

Discuss the research

Moonjump Research & Development · MotionMatrix position paper, revision 1.1 · 13 July 2026. This living document will evolve as experiments progress and evidence becomes appropriate to disclose.