H-Jiang-98/isam

iSAM is an optimization library for sparse nonlinear problems as encountered in simultaneous localization and mapping (SLAM).

Incremental Smoothing and Mapping (iSAM) library
Michael Kaess, Hordur Johannsson, David Rosen, John Leonard, 2012

isamlib/  - Source code for the iSAM library
include/  - Header files for the iSAM library
isam/     - Source code for main iSAM executable
examples/ - Example code for iSAM
doc/      - Documentation (after calling "make doc")
misc/     - Code referenced from publications
data/     - Example data files for 2D and 3D
lib/      - iSAM library (after calling "make")
bin/      - Executables (after calling "make")

Type "make doc" to generate the full source documentation including
installation instructions and examples. Then point your browser to:
doc/html/index.html

The latest documentation is also available online at:
http://people.csail.mit.edu/kaess/isam/

--------------

Overview:

The iSAM library provides a range of existing functionality for
least-squares optimization, focused on the SLAM problem (e.g. 2D/3D
pose graph, landmarks, visual SLAM). In addition to standard batch
optimization, efficient incremental optimization is provided for sets
of variables (nodes) and constraints (factors) that grow over
time. The library performs (incremental) QR matrix factorization to
solve the normal equations. Nonlinear constraints are dealt with using
Gauss-Newton, Powell's Dog leg or, for batch only, the
Levenberg-Marquardt algorithm. The default quadratic cost function can
be replaced by a general cost function, allowing the use of robust
estimators.

Beyond the provided functionality, the library can easily be extended
by the user to other sparse least-squares problems. In particular,
the user can define new nodes, containing variables to be estimated as
well as an exponential map for dealing with over-parameterized
representations. Similarly, the user can define new factors,
containing the error function for a given constraint and optionally a
symbolic Jacobian for cases in which the provided numerical
differentiation is not sufficiently fast.