Distributed Adaptive Control: An Ideal Cognitive Architecture Candidate
159
thus complementing the primary reactive navigation with adaptive responses. Using the
internal representation of space fostered by the Reactive Layer in the explorative trials,
the Adaptive Layer allowed goal-directed navigation reducing occupancy of the arena
and the mean trajectory length and increasing the collection rate. Finally, at the late
trials, the agent ended up displaying a mostly linear trajectory from the home location
to the target and back. This linear pattern was achieved thanks to the involvement of the
Contextual Layer since it combined a robust representation of the environment and made
available the goal locations stored in the long-term memory. These achievements in a
hoarding task, by implementing DAC in a mobile robot, support a similar implementation
in the context of Industry 4.0.
4.2 Previous Implementations of DAC: Human-Robot Interaction
In the context of a recycling plant, social skills such as empathy, natural language, or
social bond formation do not precisely fit the context of collaboration between humans
and robots in an industrial plant. However, the human workers will interact with the
robots, especially the robotic grippers, as they both will be required to perform tasks
with the common goal of disassembling a device. These robots will also consider that
workers may show different skills, preferences, and even trust in robots. Perceived safety,
collaboration, adaptation to each worker and the completion of a task are critical points
for successful Human-Robot Interaction.
An example of a successful collaboration that ensures safety has been shown by [18].
Here, the authors presented a collaborative human-robot assembling task. However, the
task was restricted to the assembling of a single component, always following the same
steps, and no tool manipulation was required. In contrast, the disassembling process
requires tool manipulation and includes a variety of devices, where the disassembly
steps may differ. Thus, a more adaptive solution is needed to disassemble different
WEEEs while collaborating with the human partner successfully.
Numerous contributions to the field of HRI have been provided by DAC through
its implementation in humanoid robots. However, focusing on relevant problems for
recycling plants, addressing the anchoring problem is essential. The anchoring problem
refers to the process of creating and maintaining the link between raw data provided
by the sensors and symbolic representation processed and stored by the system. [16]
tackled this problem by defining a representation of knowledge based on the so-called
H5W problem. Thus, several entities connected by semantic links (who, how, what,
where and when) full describe the situation. In other words, the Somatic Layer is taking
the sensory input data (i.e. spatial properties of objects and agents). Subsequently, the
Adaptive Layer will translate this data into instances (i.e. unsafe situation) by providing
solutions to the H5W problem. Finally, these solutions will be compared with those
stored in the Long-Term Memory of the Contextual Layer (that previously showed good
results), so the best one will be selected.
In the context of a recycling plant, considering the limited social skills of a robotic
gripper, and at the same time that interaction with the worker is needed, we propose
two channels of communication. As most plants are considered noisy, and workers wear
protection gear, verbal communication is not preferable. For this reason, we will employ
a predefined gesture-based communication of a set of fundamental requests to the robot,
159
thus complementing the primary reactive navigation with adaptive responses. Using the
internal representation of space fostered by the Reactive Layer in the explorative trials,
the Adaptive Layer allowed goal-directed navigation reducing occupancy of the arena
and the mean trajectory length and increasing the collection rate. Finally, at the late
trials, the agent ended up displaying a mostly linear trajectory from the home location
to the target and back. This linear pattern was achieved thanks to the involvement of the
Contextual Layer since it combined a robust representation of the environment and made
available the goal locations stored in the long-term memory. These achievements in a
hoarding task, by implementing DAC in a mobile robot, support a similar implementation
in the context of Industry 4.0.
4.2 Previous Implementations of DAC: Human-Robot Interaction
In the context of a recycling plant, social skills such as empathy, natural language, or
social bond formation do not precisely fit the context of collaboration between humans
and robots in an industrial plant. However, the human workers will interact with the
robots, especially the robotic grippers, as they both will be required to perform tasks
with the common goal of disassembling a device. These robots will also consider that
workers may show different skills, preferences, and even trust in robots. Perceived safety,
collaboration, adaptation to each worker and the completion of a task are critical points
for successful Human-Robot Interaction.
An example of a successful collaboration that ensures safety has been shown by [18].
Here, the authors presented a collaborative human-robot assembling task. However, the
task was restricted to the assembling of a single component, always following the same
steps, and no tool manipulation was required. In contrast, the disassembling process
requires tool manipulation and includes a variety of devices, where the disassembly
steps may differ. Thus, a more adaptive solution is needed to disassemble different
WEEEs while collaborating with the human partner successfully.
Numerous contributions to the field of HRI have been provided by DAC through
its implementation in humanoid robots. However, focusing on relevant problems for
recycling plants, addressing the anchoring problem is essential. The anchoring problem
refers to the process of creating and maintaining the link between raw data provided
by the sensors and symbolic representation processed and stored by the system. [16]
tackled this problem by defining a representation of knowledge based on the so-called
H5W problem. Thus, several entities connected by semantic links (who, how, what,
where and when) full describe the situation. In other words, the Somatic Layer is taking
the sensory input data (i.e. spatial properties of objects and agents). Subsequently, the
Adaptive Layer will translate this data into instances (i.e. unsafe situation) by providing
solutions to the H5W problem. Finally, these solutions will be compared with those
stored in the Long-Term Memory of the Contextual Layer (that previously showed good
results), so the best one will be selected.
In the context of a recycling plant, considering the limited social skills of a robotic
gripper, and at the same time that interaction with the worker is needed, we propose
two channels of communication. As most plants are considered noisy, and workers wear
protection gear, verbal communication is not preferable. For this reason, we will employ
a predefined gesture-based communication of a set of fundamental requests to the robot,