154
O. G. Rosado and P. F. M. J. Verschure
representing the natural evolution of embedded systems, going from centralised control
systems to autonomous machines capable of communicating with each other [2]; b)
Cloud computing, that not only provide Industry 4.0 with high-performance computing
and low-cost storage but also allow system orchestration by modularisation and sharing
resources in a highly distributed way; c) IoT, working as a global network infrastructure that fully integrates identities, attributes and personalities of physical and virtual
“Things”, thanks to radio-frequency identification and wireless sensor networks [3].
A predecessor of IoT in a country scale industrial context, the Cybersyn Project,
can serve us as an example of how important is taking into account the contemporary
challenges. This project aimed to collect and transmit economic-related data in realtime to aid Chile’s governmental body to make informed decisions in a more democratic
manner [4]. Cybersyn began in 1971, however, due to technical, financial and political
circumstances met its end in 1973 with Pinochet’s dictatorship [5]. To avoid similar
failures, any project aiming to get into the Fourth Industrial Revolution must consider
contemporary challenges such as: improvement of Information and Communication
Technology infrastructures, solving the scalability problem, development of data science
and data analytics techniques as well as heterogeneous IoT-related networks. In this
work, we address one specific barrier that may hinder progress: the necessity for new
architectural models. CPS (e.g. robots) must lead with uncertainty when interacting with
the natural world (e.g. industry plant). This uncertainty is due to the changing conditions,
the variety of possibilities and the complexity that open environments offer. To tackle
this problem, an architecture approach is essential since allows the CPS to be dexterous
in different competencies while ensuring safety, security, scalability, and reliability [6].
However, current architectures are not capable of fulfilling all Industry 4.0 requirements.
In this paper, a new version of the Distributed Adaptive Control (DAC) architecture is
proposed as an ideal candidate to control an industrial plant in a recursive fashion.
Artificial Intelligence approaches have shown promising results when it comes to
agents performing simple tasks in dynamic but constrained environments. For example,
logarithmic AI solutions have demonstrated successful results (even exceeding human
performance levels) in limited domains such as Atari videogames or Go, but its implementation to solve the simplest navigation task is far to be possible. Furthermore, they
require a large amount of training in comparison to human learning [7]. In contrast to
board games, industrial plants are highly complex and heterogeneous. Robots operating
within such a plant need to be equipped with a wide range of capabilities (i.e. navigation,
motor control, human-robot interaction, etc.). Due to the complex behaviour required
to these robots, an architectural strategy fits better with the necessities. With an appropriately designed architecture able to organise the different plant-specialised modules
and information flow, the system should acquire robustness while performing various
tasks. Indeed, the challenge of creating such an architecture opens the question of what
design principles must be followed. Although control architectures can accomplish the
tasks for which they have been designed, in many cases their success is constrained to a
predictable environment, and their performance is far from the human-level efficiency
[8]. In contrast, cognitive architectures aim to build human-level artificial intelligence by
O. G. Rosado and P. F. M. J. Verschure
representing the natural evolution of embedded systems, going from centralised control
systems to autonomous machines capable of communicating with each other [2]; b)
Cloud computing, that not only provide Industry 4.0 with high-performance computing
and low-cost storage but also allow system orchestration by modularisation and sharing
resources in a highly distributed way; c) IoT, working as a global network infrastructure that fully integrates identities, attributes and personalities of physical and virtual
“Things”, thanks to radio-frequency identification and wireless sensor networks [3].
A predecessor of IoT in a country scale industrial context, the Cybersyn Project,
can serve us as an example of how important is taking into account the contemporary
challenges. This project aimed to collect and transmit economic-related data in realtime to aid Chile’s governmental body to make informed decisions in a more democratic
manner [4]. Cybersyn began in 1971, however, due to technical, financial and political
circumstances met its end in 1973 with Pinochet’s dictatorship [5]. To avoid similar
failures, any project aiming to get into the Fourth Industrial Revolution must consider
contemporary challenges such as: improvement of Information and Communication
Technology infrastructures, solving the scalability problem, development of data science
and data analytics techniques as well as heterogeneous IoT-related networks. In this
work, we address one specific barrier that may hinder progress: the necessity for new
architectural models. CPS (e.g. robots) must lead with uncertainty when interacting with
the natural world (e.g. industry plant). This uncertainty is due to the changing conditions,
the variety of possibilities and the complexity that open environments offer. To tackle
this problem, an architecture approach is essential since allows the CPS to be dexterous
in different competencies while ensuring safety, security, scalability, and reliability [6].
However, current architectures are not capable of fulfilling all Industry 4.0 requirements.
In this paper, a new version of the Distributed Adaptive Control (DAC) architecture is
proposed as an ideal candidate to control an industrial plant in a recursive fashion.
Artificial Intelligence approaches have shown promising results when it comes to
agents performing simple tasks in dynamic but constrained environments. For example,
logarithmic AI solutions have demonstrated successful results (even exceeding human
performance levels) in limited domains such as Atari videogames or Go, but its implementation to solve the simplest navigation task is far to be possible. Furthermore, they
require a large amount of training in comparison to human learning [7]. In contrast to
board games, industrial plants are highly complex and heterogeneous. Robots operating
within such a plant need to be equipped with a wide range of capabilities (i.e. navigation,
motor control, human-robot interaction, etc.). Due to the complex behaviour required
to these robots, an architectural strategy fits better with the necessities. With an appropriately designed architecture able to organise the different plant-specialised modules
and information flow, the system should acquire robustness while performing various
tasks. Indeed, the challenge of creating such an architecture opens the question of what
design principles must be followed. Although control architectures can accomplish the
tasks for which they have been designed, in many cases their success is constrained to a
predictable environment, and their performance is far from the human-level efficiency
[8]. In contrast, cognitive architectures aim to build human-level artificial intelligence by