This repository has been archived by the owner on Jun 10, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 39
/
ChangeLog
62 lines (58 loc) · 3.17 KB
/
ChangeLog
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
iSAM 1.7 (Jul 2, 2013)
- Generic Linear Constraint (GLC) added to support sparsification based
on the paper "Generic Factor-Based Node Marginalization and Edge
Sparsification for Pose-Graph SLAM" by Nicholas Carleavaris-Bianco and
Ryan Eustice, ICRA 2013. See examples/glc.cpp
- Relative parameterization for homogeneous points added (McDonald et al.,
ECMR 11 and RAS 13)
- Fixed compile problem in slam_monocular.h
- Bug fix in non-relative formulation of slam_stereo.h
- Numerically more stable quaternion conversion
- Changed stack allocations of SparseVector to heap allocations to allow
for very large vectors
- Added FindCholmod.cmake as a more portable way of finding CHOLMOD
- Added option for marking nodes as deleted. The marked nodes and factors
are then deleted during update. This is more efficient then individually
removing nodes and factors
- Added support for multiple parallel instances of covariance recovery
iSAM 1.6 (Mar 30, 2012)
- Factors now take a new Noise class instead of the square root
information matrix; existing calls are converted by wrapping the
matrix with the SqrtInf() constructor; additional options are
Covariance() and Information(), which internally convert to
square root information matrix
- added Powell's dog leg algorithm for batch (superior to
Levenberg-Marquardt), and our novel incremental dog leg
algorithm (see D. Rosen et al., ICRA 2012)
- better stopping criteria for batch Gauss-Newton (see
epsilon_abs and epsilon_rel in Properties)
- added Levenberg-Marquardt option for batch optimization
- Covariance access now through Slam.covariances; cloning allows
running covariances recovery in a separate thread
- exponential map for homogeneous points in stereo, based on Quaternions
- factors for stereo vision added (see J. McDonald et al., ECMR 2011)
- exponential map made explicit, was only implicitly used in the context
of Euler angles so far; removed access_vector construct
- rotations internally replaced by Quaternions for improved optimization
- require() bug fixed, which removed all tests for Release mode;
now split into require() (always active) and requireDebug() (only
active in Debug mode, for time critical applications in inner loops)
- simplified usage of anchor nodes, see examples/achorNodes
- header file dependencies fixed (now using CMake's include_directories)
- bug fix in covariance recovery (some off-diagonal entries were incorrectly
set to 0)
- added support for accessing arbitrary covariance entries, not just
marginal covariances (see Slam.access_covariance() and
examples/covariance.cpp)
- replaced Vector/Matrix classes by third party Eigen3 library (faster, and
provides full functionality, while Vector/Matrix only provided the
functions actively being used in iSAM)
iSAM 1.5 (Nov 12, 2010)
- code optimization, reducing execution time for the Manhattan sequence by half
- added optional CSparse-based batch factorization for comparison with CHOLMOD
- changed default properties:
+ solve in every step (instead of every 10)
+ max_iterations for batch_optimization 20 (instead of 0 = unlimited)
- added toggle (key 'c') to flip background color of 3D viewer
iSAM 1.4 (Aug 29, 2010)
- initial public release