[Estimation] Add DiscreteKalmanFilterHelper class #559
Conversation
|
@fjandrad @traversaro @MiladShafiee it would be great if you could review this PR. Thanks in advance. |
|
The Travis with Valgrind tests seems to fail with the std random number generator. However, when I run the Valgrind tests on my system, it passes without any errors. |
doc/releases/v0_12.md
Outdated
| @@ -42,6 +42,7 @@ KinDynComputations finally reached feature parity with respect to DynamicsComput | |||
| * Added a new version of `changeFixedFrame` in `SimpleLeggedOdometry`class. This can be used to set a desired homogeneous transformation for the fixed frame | |||
| * Added `bindings` for `AttitudeMahonyFilter`, `AttitudeQuaternionEKF`, `DiscreteExtendedKalmanFilterHelper` (https://github.com/robotology/idyntree/pull/522) | |||
| * Added basic tests for the Attitude Estimator classes (https://github.com/robotology/idyntree/pull/522) | |||
| * Added `DiscreteKalmanFilterHelper` class for an implementation of a discrete, linear time-invariant Kalman Filter | |||
There was a problem hiding this comment.
| * Added `DiscreteKalmanFilterHelper` class for an implementation of a discrete, linear time-invariant Kalman Filter | |
| * Added `DiscreteKalmanFilterHelper` class for an implementation of a discrete, linear time-invariant Kalman Filter (https://github.com/robotology/idyntree/pull/559) |
| const double stddev = 1.0; | ||
| std::mt19937 generator(std::random_device{}()); | ||
| auto dist = std::bind(std::normal_distribution<double>{mean, stddev}, | ||
| std::mt19937(std::random_device{}())); |
There was a problem hiding this comment.
I strongly suggest not to use actually random number as seed for the test, as the day the test will actually fail, you will never be able to reproduce the failing test. Just use an arbitrary deterministic seed for the RNG, and this will make the test deterministic.
There was a problem hiding this comment.
Avoiding non-determinism in tests is also useful if we ever want to do fuzzing, see https://llvm.org/docs/LibFuzzer.html#fuzz-target .
There was a problem hiding this comment.
Removed std::random_device and and used stdlib's srand and rand instead. Fixed in the recent commit 2a25958.
I think the fact is that in the CI system the Valgrind used is the one of Xenial, and you are using the one in Bionic (3.11 vs 3.13, see https://repology.org/project/valgrind/versions). However, I think avoiding to use |
How do you suggest we test it? |
prashanthr05
force-pushed
the
feature/KalmanFilter
branch
from
0d7234d
to
2a25958
Compare
This PR adds a DiscreteKalmanFilterHelper class which implements a discrete, linear time-invariant Kalman Filter.
The steps to construct and implement the Kalman filter is described in the documentation and also in a
KalmanFilterUnitTestimplementation.The unit test, however, tests only the implementation details of the filter and not the results and performance of the filter itself.