Backpropagation-Based Analytical Derivatives of EKF Covariance for Active Sensing
Authors:
Jonas Benhamou,
Silvère Bonnabel,
Camille Chapdelaine
Abstract:
To enhance accuracy of robot state estimation, active sensing (or perception-aware) methods seek trajectories that maximize the information gathered by the sensors. To this aim, one possibility is to seek trajectories that minimize the (estimation error) covariance matrix output by an extended Kalman filter (EKF), w.r.t. its control inputs over a given horizon. However, this is computationally dem…
▽ More
To enhance accuracy of robot state estimation, active sensing (or perception-aware) methods seek trajectories that maximize the information gathered by the sensors. To this aim, one possibility is to seek trajectories that minimize the (estimation error) covariance matrix output by an extended Kalman filter (EKF), w.r.t. its control inputs over a given horizon. However, this is computationally demanding. In this article, we derive novel backpropagation analytical formulas for the derivatives of the covariance matrices of an EKF w.r.t. all its inputs. We then leverage the obtained analytical gradients as an enabling technology to derive perception-aware optimal motion plans. Simulations validate the approach, showcasing improvements in execution time, notably over PyTorch's automatic differentiation. Experimental results on a real vehicle also support the method.
△ Less
Submitted 8 March, 2024; v1 submitted 27 February, 2024;
originally announced February 2024.
Time-frequency analysis of event-related brain recordings: Connecting power of evoked potential and inter-trial coherence
Authors:
Jonas Benhamou,
Michel Le Van Quyen,
Guillaume Marrelec
Abstract:
Objective. In neuroscience, time-frequency analysis has been used to get insight into brain rhythms from brain recordings. In event-related protocols, one applies it to investigate how the brain responds to a stimulation repeated over many trials. In this framework, three measures have been considered: the amplitude of the transform for each single trial averaged across trials, avgAMP; inter-trial…
▽ More
Objective. In neuroscience, time-frequency analysis has been used to get insight into brain rhythms from brain recordings. In event-related protocols, one applies it to investigate how the brain responds to a stimulation repeated over many trials. In this framework, three measures have been considered: the amplitude of the transform for each single trial averaged across trials, avgAMP; inter-trial phase coherence, ITC; and the power of the evoked potential transform, POWavg. These three measures are sensitive to different aspects of event-related responses, ITC and POWavg sharing a common sensitivity to phase resetting phenomena. Methods. In the present manuscript, we further investigated the connection between ITC and POWavg using theoretical calculations, a simulation study and analysis of experimental data. Results. We derived exact expressions for the relationship between POWavg and ITC in the particular case of the S-transform of an oscillatory signal. In the more general case, we showed that POWavg and ITC are connected through a relationship that roughly reads $\mathrm{POWavg} \approx \mathrm{avgAMP}^2 \times \mathrm{ITC}^2$. This result was confirmed on simulations. We finally compared the theoretical prediction with results from real data. Conclusion. We showed that POWavg and ITC are related through an approximate, simple relationship that also involves avgAMP. Significance. The presented relationship between POWavg, ITC, and avgAMP confirms previous empirical evidence and provides a novel perspective to investigate evoked brain rhythms. It may provide a significant refinement to the neuroscientific toolbox for studying evoked oscillations.
△ Less
Submitted 16 November, 2022;
originally announced November 2022.
