Designing an Artificial Umwelt: Towards Natural Human-Robot Interaction

Phase one The aim within the first year of the project is to have developed a primitive
version of the umwelt level, and begin tests using multi-modal data to explore the
development of affordance-based observations. The system must be capable of recognising objects in its environment, and then recognising their affordances. Once object affordances are inferred correctly in the umwelt level, this makes the detection of communicative affordances - such as recognition of communicative intent through speech or gesture, or eye gaze - possible, opening up the ability to recognise other agents, as well as obtain channels for communicating with other agents in its environment. At this stage, the interaction the agent has with its objects around it will also ground its understanding in these interactions, creating grounded meaning.

Phase two At this stage work can beginning on attaching the perceptuo-motor and
umwelt layers together formally using the CNS level. Depending on the success of the
umwelt to properly retrieve relevant information from the perceptuo-motors, a system of
pre-processing drivers may need to be implemented which transform the data into a form
which the umwelt level is more capable of forming representations from.

Phase three A mechanism for goal production then needs to be made, which will use
the affordances available in its environment to change its internal state. Observations of
its environment will be considered internally within the umwelt, so the system will be
able to alter the state of its environment through trying to alter internal representations
which correspond to this. This will to alter its environment through the formation and
completion of goals will take place in the conscious level, and is intended to model conscious
decision-making.

Phase four By this point the system would have a theory of mind. From here work can
be done on examining the effects of scaling the system. As the model is combined, the
affordances of the system will change depending on the modality of the agent. Work in
this area can examine the abilities of the agent to recognise goals in other agents, as well
as communicative gestures such as eye gaze.