CITEC Media

How can we integrate perception with action to allow robots to move smartly and assist humans in unexpected, every-day environments?

Future robotic systems will have to autonomously operate in unpredictable environments and to smoothly interact with humans. This requires the control of complex bodies with many degrees of freedom based on input from multimodal sensory systems.

CITEC unites research groups from biology, cognitive psychology, biomechanics, and robotics in order to endow robots with a so far unachieved level of motion intelligence, ranging from the planning and execution of goal-directed behavior to the functional recognition of objects and finally to the understanding of another agent's intentions.

What mechanisms enable artificial systems to understand and actively focus on what is important, to ignore irrelevant detail, and to share attention with humans?

In order to cope with the complexity of natural environments, a key capability of cognitive systems is to rapidly recognize what is relevant and to focus their processing resources accordingly. With regard to human-machine cooperation, the capability to infer and follow the focus of attention of the human user is a basic requirement for efficient team work. In human communication, maintaining a state of shared attention is heavily influenced by emotional and social feedback signals. Modeling similar signals with technical devices, such as animated robot heads, can create a new quality of human-machine interaction. Finally, the deeper understanding of the involved attentional and socio-emotional processes will be of importance to create systems that can exhibit team intelligence.

In order to cope with these questions, CITEC links empirical research including eye-tracking, EEG-analysis and experimental work in the behavioral sciences with engineering approaches. This combination allows the creation of artificial models to gain a better understanding of what is required to build attentive systems that aim to actively align their processing resources with their human partner. A connected goal is to make computers and robots more "social" by developing ways of making the state of such systems intuitively transparent to humans, utilizing natural channels such as facial or prosodic feedback or innovative graphical displays..

How can we coherently coordinate language, perception, and action so that cooperation between humans and technical systems takes place in natural efficiency?

Evolution has brought about human communication as the most efficient, robust, and flexible means of coordinating action between agents. In situated communication, humans may interact while exposed to time pressure, the bottleneck of the medium, and the uncertainty of available information, but nevertheless cooperatively achieve a huge variety of divergent actions.

Understanding the principles of communication as well as modeling and implementing them is a challenge that CITEC will tackle in a joint effort of linguists, psycholinguists, and computer scientists. Apart from offering a major step in adaptable, personalized human-computer interaction, this implies a more seamless integration of communicative building blocks of CIT ranging from augmented reality-based assistance devices to adaptive virtual agents in web technology.

What memory architectures enable a system to acquire, store and retrieve knowledge and to improve its capabilities through learning?

Memory joins perception with experience and allows cognitive processing to generate intelligent action. However, its technical design is still confronted with major problems. What type of memory architecture provides generality, versatility, and scalability? How can cognitivist and emergent system approaches contribute? And finally: what learning approaches enable an autonomously intelligent system to develop mental capabilities that allow it to solve unforeseen problems at the time of design?

CITEC explores multiple memory systems of biological brains in order to identify components and temporal attributes necessary and sufficient to realize a similarly organic memory function on a technical platform. It investigates advanced methods intended to establish and continuously maintain the linkage between conceptual and perceptual entities on the basis of "multimodal anchoring". As a new approach to learning, it transfers a novel teaching strategy derived from the communicative principles of parent-infant interaction. We expect a leap in the quality of the endowment of technical systems, as they become individually attuned to the specific life contexts of their human partners.