156
O. G. Rosado and P. F. M. J. Verschure
more skilled in performing device disassembly, as their robotic counterparts can only
perform partial disassembly that is not generalised to all types of devices. Nonetheless,
this partial performance still represents a relief in the arduous task of component disassembling and material. A solution to expedite this work is the development of hybrid
human-robot recycling plants where experienced workers would cooperate with specialised robots. This solution implies splitting the disassembling process into subtasks
according to the skills of both humans and robots.
A clear example of this new paradigm is the European Project HR-Recycler, where
robotic grippers and mobile robots assist in the recycling process of WEEE. Here, robots
not only perform repetitive and automated tasks, but they are endowed with autonomous
behaviour adaptive to a changing context. Thus, HR-Recycler represents a step towards
Industry 4.0 and an ideal framework to introduce new architecture models able to perform
the disassembly task even under conditions of uncertainty (like a dynamic, open space in
a hybrid human-robot recycling plant). We propose an enhanced plant where we apply
the DAC architecture at both single-agent (robots) and large-scale (plant) levels, shaping
a multi-scale recursive architecture.
3 Distributed Adaptive Control
The Distributed Adaptive Control (DAC) [11] is a theory of the principles underlying
Mind, Brain and Body Nexus. It is expressed as a robot-based neural architecture that
accounts for the stability maintained by the brain between an embodied agent and its
environment through action. DAC assumes that any agent, to act, must continuously
solve four fundamental questions, the so-called H4W problem [12]: “Why”, reflects the
motivation in terms of needs, drives and goals; “What”, accounts for the objects in the
world that actions pertain to; “Where”, represents the location of the object and the self;
and “When”, serves as a temporal reference of the actions. Additionally, a fifth question
(Who) was added, referring to the agency [13].
DAC organises the generation of behaviour horizontally across four layers of control.
The Somatic Layer defines the fundamental interface between the embodied agent and
its environment, including the needs that must be fulfilled to ensure survival. In a robotic
system, this layer accounts for its sensors and actuators and sets its predefined needs.
The Reactive Layer provides a set of unconditioned responses working as reflexes for
given unconditioned stimuli. This layer represents the first stage of the generation of
goals since the behaviours produced follow homeostatic and allostatic principles. The
Reactive Layer works on top of the Somatic Layer, gathering sensory data and providing
reflex responses through the actuators. A clear example of these reflexes is the “stop signal” triggered when a human gets close to the robot trajectory. The Adaptive Layer frees
the system from the restricted reflexive system by perceptual and behavioural learning.
It follows classical conditioning principles since the value of the sensory input is shaped
by experience, and its outputs could result in anticipation response. Thanks to this layer,
the robot can adapt its behaviour according to relevant stimuli (i.e. adjusting its security distance depending on the current scenario). Finally, the Contextual Layer allows
the generation of behavioural plans or policies based on sequential memory systems.
Sequential representations of states of the environment and the sensory-motor contingencies acquired by the agent are stored in the memory systems, allowing behavioural
O. G. Rosado and P. F. M. J. Verschure
more skilled in performing device disassembly, as their robotic counterparts can only
perform partial disassembly that is not generalised to all types of devices. Nonetheless,
this partial performance still represents a relief in the arduous task of component disassembling and material. A solution to expedite this work is the development of hybrid
human-robot recycling plants where experienced workers would cooperate with specialised robots. This solution implies splitting the disassembling process into subtasks
according to the skills of both humans and robots.
A clear example of this new paradigm is the European Project HR-Recycler, where
robotic grippers and mobile robots assist in the recycling process of WEEE. Here, robots
not only perform repetitive and automated tasks, but they are endowed with autonomous
behaviour adaptive to a changing context. Thus, HR-Recycler represents a step towards
Industry 4.0 and an ideal framework to introduce new architecture models able to perform
the disassembly task even under conditions of uncertainty (like a dynamic, open space in
a hybrid human-robot recycling plant). We propose an enhanced plant where we apply
the DAC architecture at both single-agent (robots) and large-scale (plant) levels, shaping
a multi-scale recursive architecture.
3 Distributed Adaptive Control
The Distributed Adaptive Control (DAC) [11] is a theory of the principles underlying
Mind, Brain and Body Nexus. It is expressed as a robot-based neural architecture that
accounts for the stability maintained by the brain between an embodied agent and its
environment through action. DAC assumes that any agent, to act, must continuously
solve four fundamental questions, the so-called H4W problem [12]: “Why”, reflects the
motivation in terms of needs, drives and goals; “What”, accounts for the objects in the
world that actions pertain to; “Where”, represents the location of the object and the self;
and “When”, serves as a temporal reference of the actions. Additionally, a fifth question
(Who) was added, referring to the agency [13].
DAC organises the generation of behaviour horizontally across four layers of control.
The Somatic Layer defines the fundamental interface between the embodied agent and
its environment, including the needs that must be fulfilled to ensure survival. In a robotic
system, this layer accounts for its sensors and actuators and sets its predefined needs.
The Reactive Layer provides a set of unconditioned responses working as reflexes for
given unconditioned stimuli. This layer represents the first stage of the generation of
goals since the behaviours produced follow homeostatic and allostatic principles. The
Reactive Layer works on top of the Somatic Layer, gathering sensory data and providing
reflex responses through the actuators. A clear example of these reflexes is the “stop signal” triggered when a human gets close to the robot trajectory. The Adaptive Layer frees
the system from the restricted reflexive system by perceptual and behavioural learning.
It follows classical conditioning principles since the value of the sensory input is shaped
by experience, and its outputs could result in anticipation response. Thanks to this layer,
the robot can adapt its behaviour according to relevant stimuli (i.e. adjusting its security distance depending on the current scenario). Finally, the Contextual Layer allows
the generation of behavioural plans or policies based on sequential memory systems.
Sequential representations of states of the environment and the sensory-motor contingencies acquired by the agent are stored in the memory systems, allowing behavioural