A python implementation of right invariant kalman fillter; The case includes legged robots with IMU and contact sensors.
This is my course project for COMPSCI690K in UMASS Amherst.
In this repository, I reimplemented the IEKF from The Invariant Extended Kalman filter as a stable observerlink to a website.
To run the InEFK; The data cames from gazebo simulator provided in this link. ; For the forward kinematics, we used urdf files from cassie robot to measure the length for all links to get SE(3) matrices to the hip.
python ./example.py
This part will demonstrate a 3D predicted trajectory for the cassie bipedal robot.
For the covariance matrix, we use identity matrcies to build it. The predicted curve should no obvious difference.
To get the contact estimation using neural network (we add LSTM and transformer based on the original code)
The transformer is connected manually. You can find the code in this forker repo.