Xavier Dutoit

Pruning and regularization in reservoir computing

Reservoir computing is a new paradigm for using recurrent neural network with a much simpler training method. The key idea is to use a large but fixed recurrent part as a reservoir of dynamic features and to train only the output layer to extract the desired information. We propose to study how pruning some connections from the reservoir to the output layer can help on the one hand to increase the generalization ability, in much the same way as regularization techniques do, and on the other hand to improve the implementability of reservoirs in hardware.

Robust visual tracking control system of a mobile robot based on a dual-Jacobian visual interaction model

This paper presents a novel design of a robust visual tracking control system, which consists of a visual tracking controller and a visual state estimator. This system facilitates human-robot interaction of a unicycle-modeled mobile robot equipped with a tilt camera. Based on a novel dual-Jacobian visual interaction model, a robust visual tracking controller is proposed to track a dynamic moving target. The proposed controller not only possesses some degree of robustness against the system model uncertainties, but also tracks the target without its 3D velocity information. The visual state estimator aims to estimate the optimal system state and target image velocity, which is used by the visual tracking controller. To achieve this, a self-tuning Kalman filter is proposed to estimate interesting parameters and to overcome the temporary occlusion problem. Furthermore, because the proposed method is fully working in the image space, the computational complexity and the sensor/camera modeling errors can be reduced. Experimental results validate the effectiveness of the proposed method, in terms of tracking performance, system convergence, and robustness.

Robust Mobile Robot Visual Tracking Control System Using Self-Tuning Kalman Filter

This paper presents a novel design of a robust visual tracking control system, which consists of a visual tracking controller and a visual state estimator. This system facilitates human-robot interaction of a unicycle-modeled mobile robot equipped with a tilt camera. Based on a novel dual-Jacobian visual interaction model, a dynamic motion target can be tracked using a single visual tracking controller without targets 3D velocity information. The visual state estimator aims to estimate the optimal system state and target image velocity, which is used later by the visual tracking controller. To achieve this, a self-tuning Kalman filter is proposed to estimate interesting parameters online in real-time. Further, because the proposed method is fully working in image space, the computational complexity and the sensor/camera modeling errors can be reduced. Experimental results validate the effectiveness of the proposed method, in terms of tracking performance, system convergence, and robustness.

Visual state estimation using self-tuning Kalman filter and echo state network

This paper presents a novel design of visual state estimation for an image-based tracking control system to estimate system state during visual tracking control process. The advantage of this design is that it can estimate the target status and target image velocity without using the knowledge of targets 3D motion-model information. This advantage is helpful for real-time visual tracking controller design. In order to increase the robustness against random observation noise, a neural network based self-tuning algorithm is proposed using echo state network (ESN) technique. The visual state estimator is designed by combining a Kalman filter with the ESN-based self-tuning algorithm. The performance of this estimator design has been evaluated using computer simulation. Several interesting experiments on a mobile robot validate the proposed algorithms.