Here we present you the latest release of GTSAM, GTSAM `4.0.3`.
**GTSAM will now default to using the full SE(3) exponential map, instead of using the Cayley map, which should give better convergence for most problems without performance impact.**
For details, please see the [blog post](https://gtsam.org/gtsam.org/2020/07/16/new-release-gtsam.html).
Important New Features & Bugfixes
In addition to the change in default Pose3 retract, which will now be the full exponential map, GTSAM has seen a steady stream of commits since the last release, 4.0.2, which has been there for more than 6 months. A summary of the most important issues and features is below:
* Robust noise model is ready for general usage
* It can be used to replace RANSAC for some applications
* For a gentle introduction, see [this](/2019/09/20/robust-noise-model.html) awesome tutorial by Varun Agrawal
* `CombinedImuFactor` serialization is now fixed
* The ISAM2 KITTI example has a C++ port, thanks Thomas Jespersen for the help!
* Now you can choose arbitrary MATLAB install prefix for the toolbox build
* Now you can `make python-install` to install the Python toolbox
* Now you can use the Conjugate Gradient solver in Python
* Now you can install GTSAM with `pip` if you only use the Python interface
* Added `FrobeniusFactor` and `FrobeniusWormholeFactor` for robust SFM applications
* Switched to in-place update of the diagonal Hessian in LM
* expect a 3%-5% speedup, YMMV
* The Cython wrapper now can be built on Windows :)
* Kudos `tuwuhs` for the help!
* Fixed a few memory-related bugs detected by the LLVM sanitizer
* Greatly improved stability