Robot studies on saccade-triggered visual prediction

Three robot studies on visual prediction are presented. In all of them, a visual forward model is used, which predicts the visual consequences of saccade-like camera movements. This forward model works by remapping visual information between the pre- and postsaccadic retinal images; at an abstract modeling level, this process is closely related to neurons whose visual receptive fields shift in anticipation of saccades. In the robot studies, predictive remapping is used (1) in the context of saccade adaptation, to reidentify target objects after saccades are carried out; (2) for a model of grasping, in which both fixated and non-fixated target objects are processed by the same foveal mechanism; and (3) in a computational architecture for mental imagery, which generates “gripper appearances” internally without real sensory inflow. The robotic experiments and their underlying computational models are discussed with regard to predictive remapping in the brain, transsaccadic memory, and attention. The results confirm that visual prediction is a mechanism that has to be considered in the design of artificial cognitive agents and the modeling of information processing in the human visual system.     

Human hip–ankle coordination emerging from multisensory feedback control

Human sensorimotor control involves inter-segmental coordination to cope with the complexity of a multi-segment system. The combined activation of hip and ankle muscles during upright stance represents the hip–ankle coordination. This study postulates that the coordination emerges from interactions on the sensory levels in the feedback control. The hypothesis was tested in a model-based approach that compared human experimental data with model simulations. Seven subjects were standing with eyes closed on an anterior–posterior tilting motion platform. Postural responses in terms of angular excursions of trunk and legs with respect to vertical were measured and characterized using spectral analysis. The presented control model consists of separate feedback modules for the hip and ankle joints, which exchange sensory information with each other. The feedback modules utilize sensor-derived disturbance estimates rather than ‘raw’ sensory signals. The comparison of the human data with the simulation data revealed close correspondence, suggesting that the model captures important aspects of the human sensory feedback control. For verification, the model was re-embodied in a humanoid robot that was tested in the human laboratory. The findings show that the hip–ankle coordination can be explained by interactions between the feedback control modules of the hip and ankle joints.     

Grasping of extrafoveal targets: A robotic model

We present a computational model of grasping of non-fixated (extrafoveal) target objects which is implemented on a robot setup, consisting of a robot arm with cameras and gripper. This model is based on the premotor theory of attention (Rizzolatti et al., 1994) which states that spatial attention is a consequence of the preparation of goal-directed, spatially coded movements (especially saccadic eye movements). In our model, we add the hypothesis that saccade planning is accompanied by the prediction of the retinal images after the saccade. The foveal region of these predicted images can be used to determine the orientation and shape of objects at the target location of the attention shift. This information is necessary for precise grasping. Our model consists of a saccade controller for target fixation, a visual forward model for the prediction of retinal images, and an arm controller which generates arm postures for grasping. We compare the precision of the robotic model in different task conditions, among them grasping (1) towards fixated target objects using the actual retinal images, (2) towards non-fixated target objects using visual prediction, and (3) towards non-fixated target objects without visual prediction. The first and second setting result in good grasping performance, while the third setting causes considerable errors of the gripper orientation, demonstrating that visual prediction might be an important component of eye–hand coordination. Finally, based on the present study we argue that the use of robots is a valuable research methodology within psychology.     

LARES: An AI-based teleassistance system for emergency home monitoring

The latest progresses in medicine are helping people live longer and better. An ageing population is a sign of a developed society with an advanced health care system. Improved life expectancy should be welcomed as a major achievement, but it should not cause a financial or social burden. In this scenario, it is critical to support older and handicap adults to continue living independently and retaining their current lifestyle. New technological advances in Wireless Sensors Networks (WSN) and Artificial Intelligence (AI) can facilitate this task.In this direction we present lares, an AI-based system that integrates a (i) WSN for receiving information of the environment and the dependent person, (ii) an autonomous robot able to take decisions based on the received information, and (iii) a Web-based system to provide telecare assistance. lares has been tried in two dependent elderly home environments during several weeks, and the experiments show that is able to detect anomalies and generate alarms in abnormal situations.