Naeem Assif Mirza

Grounded Sensorimotor Interaction Histories in an Information Theoretic Metric Space for Robot Ontogeny

We motivate and present a definition of an embodied, grounded individual sensorimotor interaction history, which captures the time-extended behavior characteristics of humans and many animals. We present an architecture that connects temporally extended individual experience with capacity for action, whereby a robot can develop over ontogeny through interaction. Central to this is an information theoretic metric space of sensorimotor experience, which is dynamically constructed and reconstructed as the robot acts. We present results of robotic experiments that establish the predictive efficacy of the space and we show the robot developing the capacity to play the simple interaction game “peekaboo.” A quantitative investigation of the appropriate horizon length of experience for the game reveals the relationship between the length of experience and the cycle time of interaction, and suggests the importance of multiple, and possibly self-adaptive, horizon lengths.

Developing social action capabilities in a humanoid robot using an interaction history architecture

We present experimental results for the humanoid robot Kaspar2 engaging in a simple ldquopeekaboordquo interaction game with a human partner. The robot develops the capability to engage in the game by using its history of interactions coupled with audio and visual feedback from the interaction partner to continually generate increasingly appropriate behaviour. The robot also uses facial expressions to feedback its level of reward to the partner. The results support the hypothesis that reinforcement of time-extended experiences through interaction allows a robot to act appropriately in an interaction.

Using temporal information distance to locate sensorimotor experience in a metric space

Information distance is used to measure how similar sensorimotor experience is to past experience within a certain temporal horizon. Applied to groups of sensors this gives a mathematical metric on sensorimotor experience over time. We show that for complex data from a robot, large scale similarity of experience can be discovered from the robot perspective, providing a means of building an experiential interaction history

Development via information self-structuring of sensorimotor experience and interaction

We describe how current work in Artificial Intelligence is using rigorous tools from information theory, namely information distance and experience distance to organize the self-structuring of sensorimotor perception, motor control, and experiential episodes with extended temporal horizon. Experience is operationalized from an embodied agents own perspective as the flow of values taken by its sensors and effectors (and possibly other internal variables) over a temporal window. Such methods allow an embodied agent to acquire the sensorimotor fields and control structure of its own body, and are being applied to pursue autonomous scaffolded proximal development in the zone between the familiar experience and the unknown.

Extending the Temporal Horizon of Autonomous Robots

We introduce mathematically rigorous metrics on agent experiences having various temporal horizons. Sensorimotor variables accessible to the agent are treated as information-theoretic sources modelled as random variables. The time series from the sensorimotor variables over a given temporal region for different behavioural contexts ground an agent-based view of the agent’s own experiences, and the information-theoretic differences between sensorimotor experiences induce a metric space structure on the set of the agent’s possible experiences. This could allow an autonomous mobile robot to locate and navigate between its sensorimotor experiences on a geometric landscape (an experiential metric space) whose points are its possible experiences of a given temporal scope and in which nearby points are similar experiences.