LaserTitan Activity List

MCPTAM Husky Demo - Stan

MCPTAM Improvements

• Saba:
  • Back to figuring out the root cause for delay in map update
  • Make CPER default option.

pySE3

• Abdel:
  • Make it a repo and add instructions, issue remaining (gradients)

LaserTitan Front End - Melissa / Jason

• Steve/Jason:

  • Reading some papers on how to evaluate features, and define top three labs to benchmark against for feature tracking.

• Jason: Collected datasets, refined test code

  • Run OrbSLAM on live camera data
  • Add klt tracking as one of the options
  • Fully understand the patch warping and matching classes
  • Test fast thresholding method as a parameter static, adaptive
  • Define persistence metrics for evaluation, run them on multiple feature/descriptor types.
  • Fiddle with Adam's adaptive thresholding to see what can be modified to improve tracking in mcptam

LaserTitan Back End - Arun

• Arun: Taylor factor Jacobians for pose and point updates.
  • Finalize implementation of derived class, try out with xyz parametrization
  • Next up will be to require another derived class for spherical, relative factors
  • Ceres run, have a look

LaserTitan Tracker - Saba / Mohamed

• Saba:

  • Resolve remaining issues with prediction
  • Test it all out on the dataset with the covariance estimates
  • Confirm that filter equations are defined for ECEF for position/velocity, confirm coordinates for measurements
  • Convert the dataset to NED from LLA, based on the first dataset point as a reference point
  • Complete EKF refactor, in particular the base filter

• Mohamed: IMU only calibration, IMU-camera calibration project initiated.

  • Summarize Taylor camera model
  • Run kalibr tool
  • Formulate error estimation method for intrinsics to enable online intrinsic estimation

NCFRN Dataset

• Arun: Calibration for ground truth on the Waterloo test set to get calibration for GPS to camera.
  • Compare wander angles.
  • Work on alignment of postprocessing outputs with MCPTAM solution.
• Jason: Finalizing the post processing work on the dataset
  • Use the calibration data to simultaneously plot ground truth and MCPTAM solution
  • Fix the output from the post processed solution to be easy to import.
  • Evaluate current dataset collection pipeline and look for efficiency gains.
  • Add automatic evaluation of localization error relative to bag if ground truth is available.