Design Motion Capture Noise Reduction Planning & Solution

Traditional finite impulse response (FIR) or infinite impulse response (IIR) low-pass filters operate by averaging sequential historical time samples. While this effectively dampens high-frequency jitter, it inevitably introduces a phase delay directly proportional to the filter window size (often exceeding 50ms to 120ms). In humanoid robot teleoperation and closed-loop control, phase delay shifts the center-of-mass (CoM) feedback out of sync with actual physical inertia, resulting in catastrophic actuator resonance, foot slippage, or structural tip-over. Virdyn resolves this via minimum-phase predictive Error-State Kalman Filters that isolate noise within the error-state coordinate frame, suppressing jitter while keeping latency below 8.2ms.

Industrial facilities contain immense ferrous density (concrete reinforcing steel, high-power servomotors, industrial high-voltage switchgear), which warps local magnetic field lines and renders traditional 9-axis magnetometer headings useless. Virdyn deploys an adaptive magnetometer trust algorithm: when localized magnetic flux density or dip angle deviates from the pre-calibrated World Magnetic Model (WMM) threshold, the solver immediately drops magnetometer trust to zero and seamlessly transitions to a drift-free 6-axis kinematic integration mode. Global orientation stability is then sustained through kinematic body constraints and optical reference anchor points.

Foot sliding occurs because numerical integration of accelerometer noise continuously accumulates velocity errors even when the performer's foot is firmly planted on the ground. Virdyn eliminates foot sliding using a dual-stage Zero-Velocity Update (ZUPT) and contact graph pipeline. First, high-frequency IMU telemetry from foot and ankle nodes evaluates linear acceleration variance and angular velocity magnitudes to classify contact state with 99.4% certainty. Once contact is detected, the linear velocity constraint $v(t) = 0$ is dynamically injected into the Kalman filter state vector, resetting integration drift at every single step cycle.

Yes. Through VDRobot Studio and native UDP/TCP streaming protocols, Virdyn provides native ROS2 nodes and C++/Python APIs delivering clean, pre-filtered skeletal kinematics, joint angular velocities, and root Cartesian coordinates. The output formats seamlessly interface with NVIDIA Isaac Gym, Isaac Sim, Mujoco, and PyBullet for reinforcement learning policy pre-training, egocentric imitation learning datasets, and hardware-in-the-loop (HIL) teleoperation without requiring intermediate VFX clean-up passes.

Soft-Tissue Artifacts (STA) occur when muscle contraction, inertial soft tissue oscillation, and skin deformation cause wearable sensor nodes to move relative to the underlying rigid skeletal bone. Virdyn counters STA through a dual approach: mechanically, our custom elastic Lycra suits and compressive strap harnesses provide multi-point mechanical pre-tensioning to anchor nodes directly over low-adipose bony landmarks. Algorithmically, our solver deploys a dynamic model of skin-fascia compliance that treats node linkages as damped spring-mass systems, mathematically filtering out non-rigid tissue oscillations above 4.5 Hz.

Inertial systems suffer primarily from low-frequency integration drift and magnetic disturbance, but they are completely immune to visual occlusion, field-of-view limits, and ambient lighting shifts. Markerless optical systems experience zero global coordinate drift (since 3D cameras track absolute global space), but exhibit higher stochastic frame-to-frame joint jitter caused by neural network bounding-box ambiguity, visual self-occlusion (e.g., crossing arms), and motion blur. By pairing VDSuit inertial suits with optical reference cameras in a hybrid cluster, Virdyn leverages the strengths of both modalities: optical cameras bound long-term spatial drift, while inertial IMUs provide high-frequency acceleration fidelity and occluded-limb tracking.