160
O. G. Rosado and P. F. M. J. Verschure
such as start, stop, take rest position, etc., by providing the workbench with a computer
vision system able to solve the anchoring problem. Gestures, combined with a multitouch
interactive tablet, will provide more complex interaction scenarios.
4.3 Previous Implementations of DAC: Motor Control
Simple behaviours such as grabbing a tool, unscrewing a bolt or extracting and placing a
component during the disassembling task are complex movements that require lengthy
training sessions until a robotic gripper can adequately perform such actions.
Although a complete version of DAC has not been applied to control the specific
behaviours of a robotic gripper, previous studies have validated the implementation of
the DAC architecture for motor control. More specifically, motor control was achieved
with the acquisition of affordances, namely the categorisation of goal-relevant properties
of objects [19]. Within the context of DAC, Sanchez-Fibla, Duff & Verschure [20]
proposed the notion of affordance gradients: object-centred representations that describe
the consequences that an action may have on this particular object. Through objectcentred force fields, the agent was not just able to predict the outcomes of an action,
but also to generalise predictions to actions that the agent has not previously perform.
These affordance gradients were acquired through learning in Adaptive Layer, allowing
to a mobile robot to push an object from the right side and place it in a target position
and orientation. These affordance gradients were recently extended to the acquisition of
bimanual affordances in Sanchez-Fibla et al. [21].
5 DAC at the Large-Scale Level
Interestingly, DAC has also shown its capabilities to control an entertainment space.
Ada [17] was a large-scale intelligent and interactive environment that was able not
just to learn information from its visitors, but also to modify its behaviour guiding their
steps toward a given direction. Ada achieved interaction with its visitors by expressing
its internal states through global lighting and background sound. Information processing through DAC allowed leveraging multi-modal data from Ada’s sensors (cameras,
microphones and pressure-sensitive floor) to learn the best way to interact with its visitors following paradigms of classical and operant condition, demonstrating that DAC is
not constrained to conventional robots.
6 Multi-scale DAC: A Micro-Recycling Plant
For a recursive multi-scale implementation of DAC architecture, a central system is
endowed with DAC operating on a large-scale level, and so controlling the synergic
functioning of the single-agent level (Fig. 1). DAC at this large-scale level is implemented
more abstractly, since it leverages sensors and effectors of the single-agent level, leading
to an intertwined recursive multi-scale architecture. The fact that this central system
integrates information from all single agents allows new perceptions such as the amount
of aCell that are free or taken, space occupied by mobile robots or mean amount of
WEEE disassembled.
O. G. Rosado and P. F. M. J. Verschure
such as start, stop, take rest position, etc., by providing the workbench with a computer
vision system able to solve the anchoring problem. Gestures, combined with a multitouch
interactive tablet, will provide more complex interaction scenarios.
4.3 Previous Implementations of DAC: Motor Control
Simple behaviours such as grabbing a tool, unscrewing a bolt or extracting and placing a
component during the disassembling task are complex movements that require lengthy
training sessions until a robotic gripper can adequately perform such actions.
Although a complete version of DAC has not been applied to control the specific
behaviours of a robotic gripper, previous studies have validated the implementation of
the DAC architecture for motor control. More specifically, motor control was achieved
with the acquisition of affordances, namely the categorisation of goal-relevant properties
of objects [19]. Within the context of DAC, Sanchez-Fibla, Duff & Verschure [20]
proposed the notion of affordance gradients: object-centred representations that describe
the consequences that an action may have on this particular object. Through objectcentred force fields, the agent was not just able to predict the outcomes of an action,
but also to generalise predictions to actions that the agent has not previously perform.
These affordance gradients were acquired through learning in Adaptive Layer, allowing
to a mobile robot to push an object from the right side and place it in a target position
and orientation. These affordance gradients were recently extended to the acquisition of
bimanual affordances in Sanchez-Fibla et al. [21].
5 DAC at the Large-Scale Level
Interestingly, DAC has also shown its capabilities to control an entertainment space.
Ada [17] was a large-scale intelligent and interactive environment that was able not
just to learn information from its visitors, but also to modify its behaviour guiding their
steps toward a given direction. Ada achieved interaction with its visitors by expressing
its internal states through global lighting and background sound. Information processing through DAC allowed leveraging multi-modal data from Ada’s sensors (cameras,
microphones and pressure-sensitive floor) to learn the best way to interact with its visitors following paradigms of classical and operant condition, demonstrating that DAC is
not constrained to conventional robots.
6 Multi-scale DAC: A Micro-Recycling Plant
For a recursive multi-scale implementation of DAC architecture, a central system is
endowed with DAC operating on a large-scale level, and so controlling the synergic
functioning of the single-agent level (Fig. 1). DAC at this large-scale level is implemented
more abstractly, since it leverages sensors and effectors of the single-agent level, leading
to an intertwined recursive multi-scale architecture. The fact that this central system
integrates information from all single agents allows new perceptions such as the amount
of aCell that are free or taken, space occupied by mobile robots or mean amount of
WEEE disassembled.