This paper presents a novel design of face tracking algorithm and visual state estimation for a mobile robot face tracking interaction control system. The advantage of this design is that it can track a user's face under several external uncertainties and estimate the system state without the knowledge about target's 3D motion-model information. This feature is helpful for the development of a real-time visual tracking control system. In order to overcome the change in skin color due to light variation, a real-time face tracking algorithm is proposed based on an adaptive skin color search method. Moreover, in order to increase the robustness against colored observation noise, a new visual state estimator is designed by combining a Kalman filter with an echo state network-based self-tuning algorithm. The performance of this estimator design has been evaluated using computer simulation. Several experiments on a mobile robot validate the proposed control system.

Reservoir Computing (RC) is increasingly being used as a conceptually simple yet powerful method for using the temporal processing of recurrent neural networks (RNN). However, because fundamental insight in the exact functionality of the reservoir is as yet still lacking, in practice there is still a lot of manual parameter tweaking or brute-force searching involved in optimizing these systems. In this contribution we aim to enhance the insights into reservoir operation, by experimentally studying the interplay of the two crucial reservoir properties, memory and non-linear mapping. For this, we introduce a novel metric which measures the deviation of the reservoir from a linear regime and use it to define different regions of dynamical behaviour. Next, we study the relationship of two important reservoir parameters, input scaling and spectral radius, on two properties of an artificial task, namely memory and non-linearity.

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.

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.

Reservoir Computing (RC) is a relatively recent architecture for using recurrent neural networks. It has shown interesting performance in a wide range of tasks despite the simple training rules. We use it here in a logistic regression (LogR) framework. Considering non-Markovian time series with a hidden variable, we show that RC can be used to estimate the transition probabilities at each time step and also to estimate the hidden variable. We also show that it outperforms classical LogR on this task. Finally, it can be used to extract invariants from a stochastic series.

Reservoir Computing is a new paradigm for using Recurrent Neural Networks which shows promising results. However, as the recurrent part is created randomly, it typically needs to be large enough to be able to capture the dynamic features of the data considered. Moreover, this random creation is still lacking a strong methodology. We propose to study how pruning some connections from the reservoir to the readout can help on the one hand to increase the generalisation ability, in much the same way as regularisation techniques do, and on the other hand to improve the implementability of reservoirs in hardware. Furthermore we study the actual sub-reservoir which is kept after pruning which leads to important insights in what we have to expect from a good reservoir.

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 target's 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.

Electric wheelchairs can provide necessary help to elderly or disabled people. If the people are unable to give precise and fine steering commands, an intelligent electric wheelchair can assist the driving by implementing a collision avoidance behaviour. However, a straightforward implementation of such an algorithm is not desired as the intent of the users is not always to avoid the obstacle (if for instance they want to dock at a table or pick a book from a shelf). A possible improvement is then to provide the users with assistance adapted to their intent. The problem is then to estimate when the users actually need assistance. To solve this task, we apply here a neural network technique, Reservoir Computing, which is a new and simple yet powerful way to use Recurrent Neural Networks.

Reservoir Computing (RC) uses a randomly created recurrent neural network where only a linear readout layer is trained. In this work, RC is used for detecting complex events in autonomous robot navigation. This can be extended to robot localization based solely on sensory information. The robot thus builds an implicit map of the environment without the use of odometry data. These techniques are demonstrated in simulation on several complex and even dynamic environments.

Reservoir Computing is a new paradigm to use artificial neural networks. Despite its promising performances, it has still some drawbacks: as the reservoir is created randomly, it needs to be large enough to be able to capture all the features of the data. We propose here a method to start with a large reservoir and then reduce its size by pruning out neurons. We then apply this method on a prototypical and a real problem. Both applications show that it allows to improve the performance for a given number of neurons.

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 target's 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.

Autonomous mobile robots form an important research area due to their applicability in the real world as domestic service robots. The autonomy of a robot strongly relies on its ability to extract in- formation from the environment. A robot must also be aware of the current situation for an improved interaction with humans or other robots. However, it is very difficult to achieve robotic autonomy in a simple way. Robot tasks such as event detection, robot localization, plan execution, intent recog- nition are examples of complex tasks that can not be easily solved by standard robotic approaches. These tasks have to be performed using a limited number of sensors with low accuracy, as well as with a restricted amount of computational power. This work uses the recently emerged paradigm of Reservoir Computing (RC) (1) for grasping infor- mation in several contexts of autonomous robotic tasks. RC is a new concept for efficient handling of recurrent networks. With RC, the states of a random and high-dimensional dynamical system made of a reservoir of recurrent nodes are mapped onto the desired output via a simple linear read- out. Only the readout output layer is trained using standard linear regression techniques which are simple and fast to implement (the recurrent part or reservoir is left fixed). This work presents recent and ongoing research carried jointly by two research groups on several robotic tasks. By using RC, we can solve the tasks without any model of the environment neither of the task itself. We consider distinct levels of complexity for the robotic tasks: event detection, localization, plan recognition and intent detection (2). Event detection consists of detecting simple occurrences local in time and space. It is not a trivial task: for instance, the events of 'passing through door 1' and 'passing through door 2' can seem identical for a robot. It is very important that we are able to distinguish between very similar events. The next step is towards robot localization, in which we rather want to predict the current location of the robot based on the same kind of sensory information. We consider three different granularities for the problem of robot localization: coordinate detection in the cartesian map; location detection in a grid of small discrete areas; location detection in a more realistic environment composed of rooms of distinct sizes. The following robotic task corresponds to plan recognition. Particularly, we are interested in recognizing robot actions during a navigation task such as: the robot is navigating to goal number 1 (which means a possible robot action in a number of distinct goals). Finally, with intent detection we aim at predicting the current intent of a human. It is different from action in that we consider here the case of disabled user who can not always act according to his/her intents. For all four complexity levels, we use RC as a black-box model for learning the task, where the inputs are only distance sensors. In addition, in the context of robot localization, we further extend the experiments to map and path generation. This is achieved by using the RC network in a generative setting, which makes possible to easily extract the maps and the trajectories stored internally by the reservoir. This work aims at achieving increasing degrees of robotic autonomy by using a simple and efficient technique, namely, RC. We show that we can efficiently solve all aforementioned robotic tasks with RC. Finally, we believe that RC can be applied to a wide range of robotic tasks, enhancing the robot's autonomy, its interaction with (disabled) humans or other cooperating robots.