State Estimation Performance

IHMC Estimation

Yaw Rate Bias

• example bias estimation when standing still: -0.0008-0.001 rad/sec
• is quite stable when the robot is standing still (no drift and variance of +/-0.5 degrees)
• but motion of the robot (e.g. walking) can cause the bias estimate to change alot - which is clearly in the actual bias

Force Torque Tares

1. raw_force_z_leftAnkleRoll - raw value sensed by ATI sensor

2. leftFootSensorForceCalibrationOffsetZ - the calibration value taken when the tare button is pressed.

3. This seems to be the calculation adjusted contact forces: leftFootStateEstimatorForceZ = raw_force_z_leftAnkleRoll - leftFootSensorForceCalibrationOffsetZ

4. This is very close to identical, probably because the sensor and foot axes are aligned: leftFootStateEstimatorForceFootFrameZ

5. Then a straightforward magnitude is given by: leftFootStateEstimatorFootForceMag = norm( leftFootStateEstimatorForce{XYZ} )

• Encoded in WrenchBasedFootSwitch.java.
• NOTE: both the controller and the state estimator WrenchBasedFootSwitch's . The controller version seems to have more delay. The SE version is less heavily filtered