Rigid body dynamics estimation by unscented filtering pose estimation neural networks
T Avant, KA Morgansen - 2020 American Control Conference …, 2020 - ieeexplore.ieee.org
2020 American Control Conference (ACC), 2020•ieeexplore.ieee.org
In this paper, we consider the task of estimating the state of dynamic object by applying an
unscented filter to pose estimates generated by a neural network. To incorporate the
rotational state of the system into the filter, we use a parameterization of the tangent space of
the group of rotation matrices SO (3). We then characterize the noise in the pose estimation
neural network by considering simple motions of the object, as well as using a Monte Carlo
approach. Finally, using synthetically generated images, we show in simulation how the …
unscented filter to pose estimates generated by a neural network. To incorporate the
rotational state of the system into the filter, we use a parameterization of the tangent space of
the group of rotation matrices SO (3). We then characterize the noise in the pose estimation
neural network by considering simple motions of the object, as well as using a Monte Carlo
approach. Finally, using synthetically generated images, we show in simulation how the …
In this paper, we consider the task of estimating the state of dynamic object by applying an unscented filter to pose estimates generated by a neural network. To incorporate the rotational state of the system into the filter, we use a parameterization of the tangent space of the group of rotation matrices SO(3). We then characterize the noise in the pose estimation neural network by considering simple motions of the object, as well as using a Monte Carlo approach. Finally, using synthetically generated images, we show in simulation how the unscented filter can improve the accuracy of the pose estimates from the neural network.
ieeexplore.ieee.org
Showing the best result for this search. See all results