Sliding Autonomy for a Human Centered System in a Space Exploration Domain

1.1 Robots 2
• ”Introduction” describes the state of the art in human-robot interac-
tion with reference to space exploration domains;
• ”Sliding Autonomy” examines closely the main issues to take into ac-
count in developing a sliding autonomy system;
• ”Existing Frameworks for Sliding Autonomy” compares two existing
frameworks under the perspective of sliding autonomy concepts;
• chapters 4, 5, 6, 7 describe the developed software;
• ”Discussion and Conclusion” contains remarks about the competences
acquired during this work and suggestions about possible improve-
ments.
1.1 Robots
Robots integrate ideas from information technology with physical embodi-
ment. The Robot Institute of America (1979) defines a robot as a ”repro-
grammable, multifunctional manipulator designed to move materials, parts,
tools, or specialized devices through various programmed motions for the
performance of a variety of tasks”. This is only one of the possible defi-
nitions, another perspective was provided by the Webster dictionary, that
defined a robot as ”an automatic device that performs functions normally
ascribed to humans or a machine in the form of a human”.
The most important utilization of robots is in the industrial sector, where
companies exploit their performances in tasks such as assembly and trans-
portation. Usually, they do not need complex sensing and computing capa-
bilities, because they are only used in keeping of repeating always the same
tasks. Considering the huge progresses done in this field of research and due
to the massive cost reduction, it is desirable in the future to develop robots
able to provide people with direct assistance, even at home. In this con-
text, advancements in robotic autonomy allow their utilization also within
unstructured and uncertain environments, such us an employee’s workspace
or home.
1.2 Human-Robot Interaction (HRI) 3
There still are a number of open challenges before obtaining that result.
On one hand, giving the robots a high degree of autonomy provides him with
the ability to arrange variations in its surroundings. Developers typically
design its set of capabilities in order to allow him to make its own decisions
according to the specific situation it is facing. But on the other hand, new
researches aim at extending the robot capabilities through interactions with
one or more human operators. In this way, even if a robot is not capable
of accomplishing a complex job, a human can intervene, guiding him more
or less tightly in performing the right operations and hence in reaching the
goal. Many studies refer to this concept as sliding or adjustable autonomy,
and it is considered a promising way to realize ”social or sociable” systems.
1.2 Human-Robot Interaction (HRI)
The concept of human-robot interaction has only become possible in the
last decade, due to the progresses made in the field of robotics, in terms
of perception, reasoning and programming, that were the starting point for
the development of semi-autonomous system. Human-computer interaction
(HCI) is the term used to denote a computer application whose objects
and files are controlled through a software. HRI goes beyond the only tele-
operation, allowing the robot responding to extremely precise commands
from a human, about for example adjustment of a control arm, or supply-
ing the robot a path to follow. Fong , Thorpe and Bauer [51] notice that
HRI differs from HCI (Human-Computer Interaction) because it concerns
systems which have complex, dynamic control module, exhibit autonomy
and cognition, and operate in changing, real-world environments. There
have been identified some key issues to meet for the realization of HCI sys-
tems: there must be efficient ways for a human controller to interact with
semi-autonomous machines, whose interface and interaction depend on the
functions being performed. Kidd (1992) suggested that human-centered de-
sign of human-robot interaction needs to look beyond technology issues and
consider problems such as tasks allocation between the actors and safety.
This leads to the development of a system where both parties’ skills are
taken into account adding more sophisticated capabilities and robustness.
1.2 Human-Robot Interaction (HRI) 4
Human-robot cooperation involves necessarily a software module in the
system that must be used by a human. This means that the system must be
designed to interact with people intelligently. In contrast with the traditional
”black box” autonomous system, that executes prewritten sequences of low
level commands, a human-centered autonomous system recognizes people as
intelligent agents it must (remotely) communicate with. For many years in
the past one of the main aims of the researchers was to provide systems with
a high level of intelligent automation, trying to give them the best function-
alities without any interferences by users. This approach arises many issues
in domains where systems must work in situations where safety is critical
and knowledge is not enough to cover all the possibilities. This is the reason
why recently many researches address the problem under a different point of
view: not only full autonomous systems, but cooperation between humans
and robots to achieve a desired goal.
Scholtz [48] defines a set of roles a human can take in a HRI system. The
supervisor role involves monitoring and controlling of the overall situation,
evaluating it with respect to the goal. He can step in the execution loop mod-
ifying plans or providing a custom action. Operator interaction deals mainly
with modification of internal software or models when the robot behaviour
is not acceptable. As the word suggests, the mechanical interaction refers
to physical intervention that causes the desired effects on the behaviour. A
system has a peer interaction mainly if it is a multiagent platform, where
multiple humans and robots communicate each other in high level terms
to achieve a common goal. The final role is that of the bystander, who
commonly has only a subset of the available actions, for example stop the
execution to start a teleoperation session, without the possibility to directly
interact at the goal level1.
It is quite clear the robot cannot operate only in full automation mode
to let any type of interaction; instead the robot has to operate at different
degrees of automation, from full autonomy to teleoperation. Human-robot
interaction cannot be studied without consideration of a robot degree of au-
1Later in this thesis supervisor and operator will be used without distinction, also
including the bystander meaning.
1.2 Human-Robot Interaction (HRI) 5
tonomy, since it is a determining factor with regards to the tasks a robot can
perform, and the level at which the interactions take place. Typically, the
level of autonomy of a system is determined when the system is designed,
but the user supervisor can choose to run it either autonomously or manu-
ally, or choose between levels. Adjustable autonomy describes the property
of an autonomous system to change its level while the system is running.
This change can be done by a human operator, or by the autonomous sys-
tem itself. In second case, the system must reason also about which level
of autonomy is better for the actual situation, but not only: it could learn
how to operate reasoning about past interactions with a human operator on
akin situations, deciding accordingly the right degree of autonomy. Levels
can be different from each others in some different manners, such as about
the set of commands the robot executes, or about how many modules are
being autonomously or manually controlled, under what circumstances the
system will override manual control, the duration of operations, etc.
Many researches are focused in allowing human cooperation with robots.
The COBOT project [43] [44] tries to give better guidance support to pas-
sive robots, designed with the main aim to be collaborative with human
operators. The operator does not have to provide fine guidance control: the
human issues the force input while the system steers the mechanism into
the right place.
NASA’s ASRO [46] aims at developing a mobile robot to assist humans by
carrying tools, helping to manipulate objects and providing sensor informa-
tion. While the robot is physically working alongside the astronaut it is
teleoperated by a remote operator who communicates with the astronaut.
Fong et al. [45] develop a system where the robot and the user participate
in a dialogue: the first can ask the operator for help with localization or
to clarify sensor readings, while the second can provide the robot with any
lacking state information.
From the point of view of a space exploration domain, it is clear that a suc-
cessfully mission that accomplishes scientific objectives will require advanced
autonomy, but constant monitoring, to both increase safety and reliability
of the mission and decrease operation time and costs. As an esemplification
1.3 Robots in Space Exploration 6
we can consider that the energy management is critical in space exploration:
using energy for one purpose may mean that it will not be available for an-
other important task at the same time or later. People will not want to
have to calculate all the ramification of turning on an operation every time,
to be sure to avoid undesirable effects, therefore sophisticated autonomous
systems have to continually manage such resources and plan the operations
in such a way there is no conflicts. But on the other hand it is impossible
to predict what will be encountered during a space exploration mission. It
is critical that a human intervention is supported, mainly because it can
provide many benefits to the mission, for example increasing the system ca-
pabilities and reliability.
1.3 Robots in Space Exploration
A space exploration task can be performed by robots in order to achieve
geological information of an unknown environment. NASA, ESA and other
private companies are interested in the role of autonomous agents in their
space exploration missions.
We can consider a Mars exploration rover as an example of a complicate
situation where a system may operate.
On Mars, the round trip time for a radio signal to Earth is not negligi-
ble, it can be up to 40 minutes, so it is too long to think about a complete
teleoperation from Earth for lengthy missions, especially in situations where
the rover is travelling long distances over relatively uninteresting and benign
terrain and therefore the human operator should not be needed to contin-
ually teleoperate it. Hence the need for autonomy. Nevertheless, complete
automation does not solve all the problems an exploration could arise. When
the rover is travelling, it is possible, or even desirable, it meets places that
would be interesting to observe in detail. In such cases, even if a rover
could reliably execute complex exploration or analysis tasks, sometimes sci-
entists could want to have a more direct control in order to make custom
observations and tasks. When this happens, the rover autonomous system
should check that the operator does not try to perform something by acci-
1.3 Robots in Space Exploration 7
dent: the rover should let the human supervisor take control of only some
subsystems, but guarantee at the same time its safety, in order not to reach
catastrophic scenarios where the overall mission would be unavoidably com-
promised [6]. This is not to say that the robot has the final authority (it
should be against the second Asimov’s robotics rule) but rather it follows a
higher level strategy set by the human (that for example includes the robot
auto-preservation) with some sort of freedom in execution.
Another relevant aspect that contributes to the importance of an in-
teracting human-robot adjustable autonomy system is that it is difficult to
design a high-performance completely autonomous robot: it may require
too much power in terms of computer resources to handle the myriad of
situations that a rover is unlikely to encounter. Instead, it is more simple
to design an autonomous control system that cannot solve all the problems
on its own, but can ask a operator for help when it is not sure about what
to do next. Sliding autonomy supports these capabilities. See [2], [3], [4],
[5], [6], [9], [13], [21], [32], [39], [40] for more details.
1.3.1 State of the Art in Space Exploration: NASA’s MER
Missions
The state of the art in NASA robotic space exploration is the Mars Explo-
ration Rover (MER) mission.
The MER-A rover, friendly called Spirit, is the first rover that reached
Mars on January 2004. It was designed to be working for at least ninety
days, instead it goes beyond expectations celebrating the 5th year of activity
on January 2009. Opportunity, technically called MER-B rover, landed on
the Red Planet 20 days after its brother but on the opposite part of Mars, to
cover another terrain of exploration. Spirit and Opportunity combined have
logged over 10 km and operated for over 1200 sols (martian days). This is
achieved by assigning a large team of highly skilled professionals to down-
load telemetry and imagery, interpret the data, plan the rover activities,
and program and upload commands every sol, in addition to a large science
team to select science targets and tasks.