Citation Context

...tate vector ˆx B , composed of both the actual vehicle state estimation at time k and the rest of previously stored positions, the smoothing is performed by augmenting the state with a copy (or clone =-=[14]-=-) of the actual position estimate ˆη B and placing it immediately below the vehicle state ˆx B k : ⎡ ˆx B k = ⎢ ⎣ ˆη B k ˆν B k ˆη B k . ˆη B 1 ⎤ ⎥ ⎦ ⎡ P B k = ⎢ ⎣ P B η,k P B νη,k P B η,k . P B η,1,k...

Citation Context

...window [35]. Finally, for OFs, we employ the MSC-KF approach [25] to impose a pose update constraining all the views from which the feature was seen. To accomplish this, we utilize stochastic cloning =-=[29]-=- over a window of m camera poses. 4 Observability-Constrained VINS Using the VINS system model presented above, we hereafter describe how the system observability properties influence estimator consis...

Citation Context

...e state estimate at tk. This process enables the filter to explicitly account for the correlations of the vehicle state estimates between the time instants that e.g., two sets of images were recorded =-=[25]-=-. In what follows, we assume that at time tk the vehicle is at position G p(tk) = p1 with quaternion attitude 1 G q(tk) = q1, and after m steps it has moved to position G p(tk+m) = p2 with attitude 2 ...

Citation Context

...s, and ii) the fusion of proprioceptive sensor information with relative-state measurements. For the latter case, an efficient EKF-based estimation algorithm is proposed, that extends the approach of =-=[1]-=-. Extensive simulation and experimental results are presented, that verify the validity of the presented method. I. INTRODUCTION Accurate localization is a prerequisite for a robot to be able to inter...

Citation Context

...timate the features as part of the state vector. We hereafter describe how to apply our OC-VINS methodology to the MSC-KF. Each time the camera records an image, the MSC-KF creates a stochastic clone =-=[24]-=- of the sensor pose. This enables the MSC-KF to utilize delayed image measurements; in particular, it allows all of the observations of a given feature fi to be processed during a single update step (...

Citation Context

... derived from these can be integrated over time to provide pose estimates [3], or combined with proprioceptive sensory input in order to benefit from both available sources of positioning information =-=[9]-=-, [10]. This paper focuses on how to optimally implement the latter approach using an extended Kalman filter (EKF) [11]. This paper does not consider the case in which the feature measurements are use...

Citation Context

...d at a rate of 100Hz, whereas the time needed for processing each image is approximately 60msec. In order to treat the existing processing delays, we employ an approach similar to the one proposed in =-=[24]-=- for treating measurements that depend on previous states. In particular, at time-step k, when an image is registered, a copy of the filter state is created and added to the state vector. The error st...

Citation Context

... )−1 BC−1 + C−1 = ( HTf M −1Hf )−1 (95) and, M = HsPssHTs + R. 9We employ stochastic cloning over m time steps to ensure that the crosscorrelations between the camera poses are properly accounted for =-=[46]-=-. During each iteration, the parameter vector is updated as x⊕ = x + [ Υ1 Υ2 ΥT2 Υ4 ][ D E ] z̃. (96) After the minimization process converges, we compute the posterior covariance of the new state (i...

Citation Context

...he kernel system. That situation is true provided that the kernel system has been representing the variant and directly observed part of the full system. A similar concept to the v.u. can be found in =-=[64]-=-. The v.u. is of the form: h(Xa) = Xa − ξ = 0 (5.19) where ξ is unknown and it does not need to be known. Similarly the quality of this v.o. does not need to be calculated, it is implicitly used but n...