Deep Learning in SLAM

Application of PoseNet and dynamic structural data generation for real-time localization

Dependencies

1. Python 2.7
2. OpenCV 3
3. TensorFlow
4. MatPlotlib/ NumPy/ urllib2
5. GTSAM 4.0 with Python 2.7 enabled

GTSAM with Posenet Factor

This is the GTSAM iSAM2 solver equipped with Posenet as a sensor model and odometry get from wheel encoder as an action model.

Attributes

1. add an odometry by calling solver.step(odometry, odometryNoise)
2. add a measurement by calling solver.addObs(measurement, measurementNoise)
3. update the graph and return the current estimate by calling solver.update()

References

1. PoseNet
2. Structural Data Generation
3. Georgia Tech Smoothing and Mapping (GTSAM)
4. NCLT Dataset