Birger Johansson

Ikaros: Building cognitive models for robots

The Ikaros project started in 2001 with the aim of developing an open infrastructure for system-level brain modeling. The system has developed into a general tool for cognitive modeling as well as robot control. Here we describe the main parts of the Ikaros system and how it has been used to implement various cognitive systems and to control a number of different robots ranging from robot arms and hands to active vision systems and mobile robots.

Anticipatory Models in Gaze Control: A Developmental Model

Infants gradually learn to predict the motion of moving targets and change from a strategy that mainly depends on saccades to one that depends on anticipatory control of smooth pursuit. A model is described that combines three types of mechanisms for gaze control that develops in a way similar to infants. Initially, gaze control is purely reactive, but as the anticipatory models become more accurate, the gain of the pursuit will increase and lead to a larger fraction of smooth eye movements. Finally, a third system learns to predict changes in target motion, which will lead to fast retuning of the parameters in the anticipatory model.

An Experimental Study of Anticipation in Simple Robot Navigation

This paper presents an experimental study using two robots. In the experiment, the robots navigated through an area with or without obstacles and had the goal to shift places with each other. Four different approaches (random, reactive, planning, anticipation) were used during the experiment and the times to accomplish the task were compared. The results indicate that the ability to anticipate the behavior of the other robot can be advantageous. However, the results also clearly show that anticipatory and planned behavior are not always better than a purely reactive strategy.

Anticipation in Attention

Although attention can be purely reactive, like when we react to an unexpected event, in most cases, attention is under deliberate control anticipating events in the world. Directing attention and preparing for action takes time, and it is thus useful to be able to predict where an important event will occur in the environment and direct attention to it even before it happens. Another reason for the need for anticipation is the processing delays in the visuomotor system. In the human system it takes at least 100 ms to detect a visual target (Lamme and Roelfsema, 2000) and to just look at a moving object, we need to anticipate its movement to control the muscles of the eyes to move our gaze to the location where the target will be (von Hofsten and Rosander, 1997). The role of anticipation in attention can also be seen in the close connection between attention and action (Balkenius, 2000).

Prediction Time in Anticipatory Systems

We investigated the role of the length of the future time interval in which an agent predicts what will happen. A number of simulated robot experiments were performed where four thieves try to collect pieces of gold from a house that is guarded by a single robot. The thieves try to anticipate the movement of the guard to select behaviors that will allow them to steel the gold without being seen. This scenario was investigated in four experiments with different visual fields of the guard and different strategies of the thieves. The results show that it is not always better to predict longer into the future and that best behavior would results when the agents match their predictions to the time it will take to perform their tasks.

A Computational Model of Pupil Dilation

ABSTRACT We present a system-level connectionist model of pupil control that includes brain regions believed to influence the size of the pupil. It includes parts of the sympathetic and parasympathetic nervous system together with the hypothalamus, amygdala, locus coeruleus, and cerebellum. Computer simulations show that the model is able to reproduce a number of important aspects of how the pupil reacts to different stimuli: (1) It reproduces the characteristic shape and latency of the light-reflex. (2) It elicits pupil dilation as a response to novel stimuli. (3) It produces pupil dilation when shown emotionally charged stimuli, and can be trained to respond to initially neutral stimuli through classical conditioning. (4) The model can learn to expect light changes for particular stimuli, such as images of the sun, and produces a “light-response” to such stimuli even when there is no change in light intensity. (5) It also reproduces the fear-inhibited light reflex effect where reactions to light increase is weaker after presentation of a conditioned stimulus that predicts punishment.

Outline of a sensory-motor perspective on intrinsically moral agents

We propose that moral behaviour of artificial agents could (and should) be intrinsically grounded in their own sensory-motor experiences. Such an ability depends critically on seven types of competencies. First, intrinsic morality should be grounded in the internal values of the robot arising from its physiology and embodiment. Second, the moral principles of robots should develop through their interactions with the environment and with other agents. Third, we claim that the dynamics of moral (or social) emotions closely follows that of other non-social emotions used in valuation and decision making. Fourth, we explain how moral emotions can be learned from the observation of others. Fifth, we argue that to assess social interaction, a robot should be able to learn about and understand responsibility and causation. Sixth, we explain how mechanisms that can learn the consequences of actions are necessary for a robot to make moral decisions. Seventh, we describe how the moral evaluation mechanisms outlined can be extended to situations where a robot should understand the goals of others. Finally, we argue that these competencies lay the foundation for robots that can feel guilt, shame and pride, that have compassion and that know how to assign responsibility and blame.

Endowing Artificial Systems with Anticipatory Capabilities: Success Cases

This book has provided various theoretical perspectives on anticipatory processes in natural and artificial cognitive systems. Advantages have been proposed and confirmed in various detailed case studies, which may have given the reader detailed insights into anticipatory processes and their importance in various cognitive systems tasks. To wrap up these advantages and give a concluding overview of various current anticipatory process advantages, this final chapter highlights a concise collection of precise success stories of anticipations in artificial cognitive systems. We survey fourteen case studies, which were developed during the EU project MindRACES. In these studies, simulated or real robots were tested in different environmental tasks, which required advanced sensorimotor and cognitive abilities. These abilities included the initiation and control of goal-directed actions, the orientation of attention, finding and reaching goal locations, and performing mental experiments for action selection. All the studies have shown advantages of anticipatory mechanisms compared to reactive mechanisms in terms of increased robot autonomy and adaptivity. In some cases, anticipations even caused the development of new cognitive abilities, which were simply impossible without anticipatory mechanisms. For each case study, we indicate relevant associated publications, in which the interested reader may find further details on the relevant computational architectures, the involved anticipatory mechanisms, as well as on the analytical and quantitative results. While the book as a whole has laid out the theoretical principles and design methodology for such advancements, this final chapter thus provides various possible starting points for further developments in both the surveyed system architectures and the presented solutions to the cognitive tasks addressed.

From Focused Thought to Reveries: A Memory System for a Conscious Robot

We introduce a memory model for robots that can account for many aspects of an inner world, ranging from object permanence, episodic memory, and planning to imagination and reveries. It is modeled after neurophysiological data and includes parts of the cerebral cortex together with models of arousal systems that are relevant for consciousness. The three central components are an identification network, a localization network, and a working memory network. Attention serves as the interface between the inner and the external world. It directs the flow of information from sensory organs to memory, as well as controlling top-down influences on perception. It also compares external sensations to internal top-down expectations. The model is tested in a number of computer simulations that illustrate how it can operate as a component in various cognitive tasks including perception, the A-not-B test, delayed matching to sample, episodic recall, and vicarious trial and error.

The Expression of Mental States in a Humanoid Robot

We explore to what degree movement together with facial features in a humanoid robot, such as eyes and mouth, can be used to convey mental states. Several animation variants were iteratively tested in a series of experiments to reach a set of five expressive states that can be reliably expressed by the robot. These expressions combine biologically motivated cues such as eye movements and pupil dilation with elements that only have a conventional significance, such as changes in eye color.