16322-3d-pose

16.322 3D Pose Estimation Project - A 3D attitude estimator based on the Unscented Kalman Filter (UKF) using a MEMS IMU and magnetometer.

This repo contains a implementation of Edgar Kraft's quaternion-based UKF and a program for running the UKF on simulated data or saved experimental data. It also contains sensor drivers for the MPU-9150 on the BeagleBone Black, quaternion utilities, and tools for sensor characterization.

Installation

Clone the repo, and its submodules (you need to include the --recursive flag to get the submodules):

git clone --recursive git@github.mit.edu:mvernacc/16322-3d-pose.git

Install pip (if you don't already have it):

sudo apt-get install python-pip

Install 16322-3d-pose using pip. This will automatically install the dependencies.

cd 16322-3d-pose
pip install -e .

If you wish to use the sensor drivers, you will need to install them too:

cd 16322-3d-pose/sensor_drivers/beaglebone-mpu9x50
ipython setup.py install develop

Usage

Estimator

The estimator program is in the top-level directory:

cd 16322-3d-pose

To run the estimator on saved experimental sensor data:

ipython run_estimator.py -- --meas_source pickle --pkl_file data/exemplars/xyz90.p --use_mag --mag_cal data/exemplars/mag_cal.p

To run the estimator on simulated data:

ipython run_estimator.py -- --meas_source sim --sim_type xyz90 --use_mag

To see all options for running the estimator:

ipython run_estimator.py -- -h

Sensor Characterization

The sensor characterization programs are in the estimators/sensor_models directory:

cd 16322-3d-pose/estimators/sensor_models

To run Allan variance analysis on the rate gyroscope:

ipython allan.py -- ../../data/exemplars/data_for_allan.p gyro_data

To run magnetometer calibration:

ipython mag_cal.py -- --meas_source pickle --pkl_file ../../data/exemplars/data_for_mag_cal.p --solver leastsq

Dependencies

• python 2.7

• numpy

• scipy

• matplotlib

• beaglebone-mpu9x50 for sensor drivers.

• beaglebone-px4flow for sensor drivers.

• estimators for Kalman Filter ans sensor models.

• transforms3d for quaternion math.