Inertial Measurement Units (IMUs) – Theory and Practice H.J. Sommer III, Ph.D. The Pennsylvania State University University Park, PA 16802 hjs1@psu.edu www.mne.psu.edu/sommer IMU Tutorial 10.05.12 1 Inertial Measurement Unit ? Kinematic measurements using inertial references Attitude and magnetic heading Angular velocity Acceleration Fuse data to provide more reliable results IMU Tutorial 10.05.12 2 Inertial Measurement Unit ? 14x28 mm IMU Tutorial 10.05.12 3 Inertial Measurement Unit ? IMU Tutorial 10.05.12 4 Inertial Measurement Unit ? qOP m, JP qOP qP h rP qOP+ qP rP r s IMU Tutorial 10.05.12 q qO q 2 5 Inertial Measurement Unit ? IMU Tutorial 10.05.12 6 Inertial Measurement Unit ? IMU Tutorial 10.05.12 7 Traditional Kinematic Measurements Photogrammetry Goniometry Absolute location of point markers Relative angles across body segments Electromagnetic digitizers 6DOF of discrete sensors IMU Tutorial 10.05.12 8 Photogrammetry IMU Tutorial 10.05.12 9 Photogrammetry Quiz (for Oldtimers) Vanguard or RightGuard? DLT or BLT? Lo-Cam or Hi-Cam? IMU Tutorial 10.05.12 10 Photogrammetry Positive Absolute location and attitude of body segments Multiple IR cameras with ambient lighting Automatic marker tracking No cables to subject > 100 Hz, high resolution Markerless motion capture (MMC) IMU Tutorial 10.05.12 11 Photogrammetry Negative Calibration relative to anatomy (joints and mass centers) Requires finite differences for velocity and acceleration Marker occlusion Soft tissue artifact Limited workspace in a gait lab IMU Tutorial 10.05.12 12 Goniometry IMU Tutorial 10.05.12 13 Goniometry Positive Direct measurement of joint motion Easy to use Negative Does not measure absolute position/attitude Physical attachment to subject IMU Tutorial 10.05.12 14 Electromagnetic Digitizers IMU Tutorial 10.05.12 15 Electromagnetic Digitizers Positive 6 DOF for each body segment Negative Limited workspace Cables (new wireless) Physical attachment to subject Accuracy degraded by speed IMU Tutorial 10.05.12 16 IMUs Integrated Kinematic Sensor (IKS) Wu and Ladin, 1993 IMU Tutorial 10.05.12 17 IMUs Attitude relative to gravity vector Magnetic heading Rotational velocity Translational acceleration IMU Tutorial 10.05.12 18 IMUs Positive Absolute attitude of body segments Direct measurement of angular velocity Direct measurement of acceleration No marker occlusion Large work space in unstructured environment IMU Tutorial 10.05.12 19 IMUs Negative Does not provide absolute location, translational velocity or rotational acceleration Calibration relative to anatomy Soft tissue artifact Data communication < 100 Hz, medium resolution IMU Tutorial 10.05.12 20 History of IMUs Vehicle navigation Intercontinental ballistic missiles (ICBM) Nuclear submarines Cruise missiles MicroElectroMechanical Systems (MEMS) Automotive Consumer products IMU Tutorial 10.05.12 21 MEMS IMUs - Automotive Automotive Accelerometers to deploy airbags Vehicle roll handling IMU Tutorial 10.05.12 22 MEMS IMUs – Consumer Products Games (WiiMote) PDA (iPhone) Camera stabilization Hard disks IMU Tutorial 10.05.12 23 MEMS Fabrication IMU Tutorial 10.05.12 24 MEMS Comb Sensor/Drive IMU Tutorial 10.05.12 25 MEMS accelerometer (proof mass) gravity acceleration IMU Tutorial 10.05.12 26 MEMS accelerometer IMU Tutorial 10.05.12 27 MEMS gyro (tuning fork) IMU Tutorial 10.05.12 28 MEMS magnetometer (magnetoresistive) IMU Tutorial 10.05.12 29 MEMS IMU Outputs Signal Analog voltage (0 to 3V) Fixed frequency, variable duty cycle Digital (internal A/D converter) Bandwidth < 150 Hz IMU Tutorial 10.05.12 30 Two-Dimensional (2D) IMU Biaxial accelerometer Uniaxial gyro IMU Tutorial 10.05.12 31 Three-Dimensional (3D) IMU Triaxial accelerometer Triaxial gyro Triaxial magnetometer Required to determine spin about gravity vector az ay ax IMU Tutorial 10.05.12 32 MEMS 9DOF IMU Triaxial accelerometer Triaxial gyro ±300 deg/sec (dps), 3.3mV/dps, 140 Hz Triaxial magnetometer ±3g, 300 mV/g, 550 Hz 50 Hz On-board CPU, serial I/O IMU Tutorial 10.05.12 33 Break Time Stand up Stretch Say hello to your neighbor IMU Tutorial 10.05.12 34 Data Fusion Sensor uncertainty Geometric Rigid body Articulated model State space Kalman filter IMU Tutorial 10.05.12 35 Sensor Uncertainty s = measured signal b = zero drift or bias (function of temp) f = scale factor (function of temp) w = Gaussian white noise s2 = variance q b f sGYRO s w IMU Tutorial 10.05.12 36 LSY530 gyro ±300 degps Nonlinearity ±1% b = 1.23 V, 0.05 degps/C° f = 300 degps/V, 0.05 %/C° s = 0.035 degps/sqrt(Hz) pink noise q b f sGYRO s w IMU Tutorial 10.05.12 37 Rigid Body Fusion ayC axC C r Multiple IMUs per body Parallel axes Rejects gravity effects q ayD axD q a a / r xD xC D IMU Tutorial 10.05.12 38 Articulated Model - Pendulum simple pendulum J m P G2 q m g P G sin q 0 P G q m g P G sin q / J m P G2 G q G ayD m PG PD a xD g sin q1 2 JG m P G uniform rod a xD 0 for P D 43 P G IMU Tutorial 10.05.12 axD D a xD PD q g sin q q a g sin q / PD xD 39 Multiple Segment Model REV q r q DR r REV q q q r REV DR q q qq DR 27 x 27 REV 18 x1 IMU Tutorial 10.05.12 27 x1 27 x1 q G2 2 A G 2 G 2 s RHL 2 / G 2 2 G3 G2 2 q G 3 A G 3 s RAN 3 / G 3 q G 2 A G 2 s RAN 2 / G 2 q G2 4 A G 4 G 4 s RKN 4 / G 4 q G2 3 A G 3 G 3 s RKN 3 / G 3 2 G5 G4 2 q G 5 A G 5 s RHP 5 / G 5 q G 4 A G 4 s RHP 4 / G 4 q G2 6 A G 6 G 6 s RWA 6 / G 6 q G2 5 A G 5 G 5 s RWA5 / G 5 q 2 A G 7 s q G2 6 A G 6 G 6 s RSL 6 / G 6 G 7 G 7 RSL 7 / G 7 q G2 8 A G 8 G 8 s REL8 / G 8 q G2 7 A G 7 G 7 s REL7 / G 7 2 G9 2 A G 8 s q A s q G 9 G 9 RWR 9 / G 9 G8 G8 RWR 8 / G 8 q G2 10 A G10 G10 s RNK 10 / G10 q G2 6 A G 6 G 6 s RNK 6 / G 6 40 Kalman Filter Uses state space model Position Velocity Adaptive time domain filter Combines states Tracks variance-covariance Rejects zero drift IMU Tutorial 10.05.12 41 Kalman Filter - 2D IMU IMU Tutorial 10.05.12 42 Kalman Filter - Simplified Measure q t and q t at time t Computeq t q t q t t / t Compareq to q t t Computeq t q t q t t Compareq t to q t IMU Tutorial 10.05.12 43 Kalman Filter – Prediction probability q latitude IMU Tutorial 10.05.12 44 Kalman Filter - Measurement probability q latitude IMU Tutorial 10.05.12 45 Kalman Filter - Correction probability q latitude IMU Tutorial 10.05.12 46 Kalman Filter - Prediction probability constant speed fixed time q latitude IMU Tutorial 10.05.12 47 Kalman Filter – 2D IMU probability qPRED qCORR q t t dynamicmodel q angle IMU Tutorial 10.05.12 48 Kalman Filter - Extended State space Nonlinear state relationships Include acceleration ax-ay-qdot versus qqdot Include geometric multisegment model Include states for multiple bodies IMU Tutorial 10.05.12 49 Kalman Filter estimatestatexk Axk 1 wk 1 based on measurements zk Hxk vk trackPk var cov of xk xk 1 prior est imat e A stat edynamicmat rix wk 1 processnoise H sensor dynamicmat rix vk sensor noise IMU Tutorial 10.05.12 Q var cov of w R var cov of v Kk adaptivegain matrix 50 Kalman Filter prediction x Axk 1 P APk AT k k correction K k P H HP H R xk xk K k zk Hxk Pk I K k HP k T T 1 k k IMU Tutorial 10.05.12 51 Applications Stationary Simple attitude Simple motion Coordinated movement Inverse dynamics IMU Tutorial 10.05.12 52 Stationary Minimal change in sensor orientation Hand/arm tremor Extended arm, tracing spiral Triaxial accelerometer, >150 Hz Postural sway Supracranial accelerometer Lumbar accelerometer IMU Tutorial 10.05.12 53 Simple Attitude Body position during sleep Treatment for sleep apnea Triaxial accelerometer, very low sample rate Not interested in spin about gravity vector Restless Leg Syndrome (RLS) Monitor sudden movement High frequency sample rate Interested in event itself, not characterization IMU Tutorial 10.05.12 54 Simple Motion Planar lifting or reaching Simple articulated model 2D IMU provides position, velocity, acceleration Passive manipulation or drop Assess spasticity Compute jerk from acceleration IMU Tutorial 10.05.12 55 Coordinated Movement Basic assessment Triaxial accelerometer, >100 Hz Number of strides, timing Asymmetry of motion Rehabilitation, prosthetic fitting Full body motion Thirteen 9DOF IMUs Multiple segment model IMU Tutorial 10.05.12 56 Inverse Dynamics 2D Lower data throughput (3ch versus 9ch) Require sagittal and frontal IMUs Does not require magnetometers 3D Lifting or reaching most promising Difficulty in assessing absolute location of feet IMU Tutorial 10.05.12 57 Practical Considerations Motion variables Number of IMUs Consider alternate signals to describe motion May require two per segment Synchronization In-shoe pressure transducers IMU Tutorial 10.05.12 58 Data Transfer Umbilical with local A/D Belt-pack data logger Belt-pack wireless SD card Bluetooth, longer battery life Network wireless Dropouts, battery life IMU Tutorial 10.05.12 59 Commercial Systems Xsens MVN Biosyn FAB NexGen Ergonomics Microstrain wireless MEMSense Sparkfun WiTilt Nintendo WiiMote IMU Tutorial 10.05.12 60