Powered Back Exoskeleton — Biomechanics & AI Platform
Real-time lumbar torque prediction and OpenSim biomechanics studio for active back-support exoskeletons.
Step 1: Data Ingestion, OpenSim Scaling & Force Assignments
Importing multi-modal sensor files into the webapp, static calibration, automated Rajagopal2015 model scaling, Inverse Kinematics, and interactive external force assignments.
Multi-modal sensor file import → automated 37-DOF Rajagopal2015 model scaling & force assignments.
Seamless loading of exported .mot / .trc simulation artifacts for full 3D skeletal kinematic inspection.
Automatic sheet matching, live multi-graph DSP filter tuning, custom deep neural builder & GPU training.
Drag-and-drop model ZIP deployment with real-time continuous torque inference vs ground truth.
Technical Subsystems
5 SubsystemsMulti-Modal Sensor Acquisition
8–22 kg symmetrical, asymmetrical & fatigue lifts — 17 wireless Xsens IMUs (60 Hz) + 8–10 Cometa sEMG lumbar electrodes (1980 Hz).
Headless OpenSim Physics Solver
Automated 37-DOF musculoskeletal scaling, inverse kinematics, and 3D ground-truth lumbar torque computation (τext, τbend, τrot).
Strictly Causal DSP Filtering
Single-pass Butterworth SOS bandpass, rolling MAD outlier rejection, and 60 Hz RMS envelope synchronization for real-time hardware.
Full-Stack Neural Studio
Multi-file dataset stacking, 100+ channel selector, modular Keras architecture builder, and live WebSocket GPU training telemetry.
Parquet Caching & Docker GPU Pipeline
Apache Parquet columnar caching, NumPy zero-copy sliding windows, and Docker Compose orchestration with NVIDIA GPU passthrough.
Interested in my machine learning & robotics thesis?
I would be glad to discuss biomechanical modeling, neural network architectures, and full-stack software design.