IEEE Robotics & Automation Magazine - September 2023 - 151

No Autonomy
Robot Assistance
Task Autonomy
Conditional Autonomy
High Autonomy
Full Autonomy
FIGURE 2. Mapping between autonomy levels and the required knowledge levels. Higher autonomy requires broader knowledge.
CS: common sense.
monitor some working variables, such
as current levels, maximum speed, and
electrical noise. The specific phases and
steps of the intervention are determined
by the surgeon, who directly controls the
surgical actions. However, even when
autonomous functions are limited (levels
0 and 1), a certain amount of commonsense
knowledge is implicitly included
in the control algorithms and data analysis
methods. For example, autonomy
level 0 has specific actions that include,
e.g., tremor suppression and maintaining
orientation during clutching. At level 1,
the robot provides dexterity and cognitive
support to the human for specific actions
or tasks but not for strategic decisions
(plans) or actions, and all common-sense
reasoning is provided by the surgeon.
When a surgical robot is able to execute
some individual actions (levels 2
and 3, becoming an ARSS), it would
require the presence of surgical common
sense. In fact, the robot would need
to perform some basic reasoning on the
patient-specific pathology and anatomy,
with the surgeon ready to intervene if
needed. Referring to Figure 1, the robot
would have to interpret " near the operative
field " and decide where to insert
the needle, given the patient's anatomy.
Furthermore, in the aforementioned
example, the robot would have to know
that a suture needs a surgical needle,
which has to be grasped with a specific
orientation for optimal insertion. It has
to control motion parameters (force and
velocity) during the insertion, and it has
to know how to close the suture. Some
concepts of medical common sense
would be needed to give the ARSS more
autonomy and connect multiple phases
of an intervention. For example, the
ARSS needs to know the human anatomy
and how each step impacts it. However,
it is not always easy to identify a
sharp division between surgical common
sense and medical
common sense as a
combination of both is
necessary to specialize
a well-defined textbook
procedure into the real
intervention.
To achieve higher
autonomy, an ARSS
must be able to make
autonomous decisions
related to the complete
procedure. It will generate,
select, execute, and
monitor a surgical plan;
adapt it to different anatomies and react
to unexpected situations. Enhanced sensing,
situational awareness, and reasoning
technologies are key to achieving such
capabilities [1], together with a broader
common sense to properly assess and
react to the situation. Although medical
common sense would allow reaching
high autonomy (level 4), fully autonomous
systems (level 5) would need the
integration of general common sense
because decisions would be required
that go outside the medical field, such as
lifting and moving objects, turning on
and off devices and lights, and understanding
human emotions.
"
TO ACHIEVE HIGHER
AUTONOMY, AN
ARSS MUST BE ABLE
TO MAKE AUTONOMOUS
DECISIONS RELATED
TO THE COMPLETE
PROCEDURE.
„
CONCLUSION
As long as surgical robots maintain an
assistive role, they can rely on the common
sense of the operating surgeons.
As soon as they become aids or surgical
colleagues, they must be able to
perform common-sense reasoning on
their own, making it crucial to understand
how to deal with
this kind of knowledge.
The map between the
common-sense types
and autonomy levels
proposed in this article
aims at making the
problem more tractable
and encourages researchers
to fill the scientific
and technological gaps
related to common-sense
knowledge and reasoning.
Data-driven deep
learning algorithms are
a promising approach to embed common
sense into a process, as data implicitly
encode common-sense knowledge.
However, this common sense is neither
explicitly formalized nor identified,
thus process reasoning is not directly
explainable to the user, and it also violates
the upcoming regulations on
using artificial intelligence methods in
high-risk applications [10]. For these
reasons, we feel that common sense
should mostly rely on explainable reasoning,
especially when used in safetycritical
systems.
(continued on page 163)
SEPTEMBER 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE
151
Level 0-1
Level 2-3
Level 4
Level 5
Procedural
Knowledge
Surgical CS
Medical CS
General CS
Sequence of Surgical Tasks
Procedure-Specific CS
General Medical CS
General CS Concepts
Field-Specific CS

IEEE Robotics & Automation Magazine - September 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2023