Towards Safe Robots: Approaching Asimov's 1st Law
post by Utopiah · 2012-08-16T08:53:21.431Z · LW · GW · Legacy · 5 commentsTowards Safe Robots: Approaching Asimov's 1st Law
http://darwin.bth.rwth-aachen.de/opus3/volltexte/2011/3826/pdf/3826.pdf (via http://www.euron.org )
Despite the title very little theory or philosophy but instead a focus on interaction (e.g. in a factory, between a human worker and a robot) and how to minimize risk: soft-robotics, crash-testing, collisions, ...
Abstract
Up to now, state-of-the-art industrial robots played the most important role
in real-world applications and more advanced, highly sensorized robots were
usually kept in lab environments and remained in a prototypical stadium. Var-
ious factors like low robustness and the lack of computing power were large
hurdles in realizing robotic systems for highly demanding tasks in e.g. do-
mestic environments or as robotic co-workers. The recent increase in techno-
logy maturity finally made it possible to realize systems of high integration,
advanced sensorial capabilities and enhanced power to cross this barrier and
merge living spaces of humans and robot workspaces to at least a certain ex-
tent.
In addition, the increasing effort various companies have invested to realize
first commercial service robotics products has made it necessary to properly
address one of the most fundamental questions of Human-Robot Interaction:
How to ensure safety in human-robot coexistence?
Although the vision of coexistence itself has always been present, very little
effort has been made to actually enforce safety requirements, or to define safety
standards up to now.
In this dissertation, the essential question about the necessary requirements
for a safe robot is addressed in depth and from various perspectives. The ap-
proach taken here focuses on the biomechanical level of injury assessment, ad-
dressing the physical evaluation of robot-human impacts and the definition of
the major factors that affect injuries during various worst-case scenarios. This
assessment is the basis for the design and exploration of various measures
to improve the safety in human-robot interaction. They range from control
schemes for collision detection, and reaction, to the investigation of novel joint
designs. An in-depth analysis of their contribution to safety in human-robot
coexistence is carried out.
In addition to this “on-contact” treatment of human-robot interaction, the the-
sis proposes and discusses real-time collision avoidance methods, i.e. how to
design pre-collision strategies to prevent unintended contact. An additional
major outcome of this thesis is the development of a concept for a robotic co-
worker and its experimental verification in an industrially relevant real-world
scenario. In this context, a control architecture that enables a behavior based
access to the robot and provides an easy to parameterize interface to the safety
capabilities of the robot was developed. In addition, the architecture was ap-
plied in various other applications that deal with physical Human-Robot In
teraction as e.g. the first continuously brain controlled robot by a tetraplegic
person or an EMG2 controlled robot.
Generally, all aspects discussed in this thesis are fully supported by a variety
of experiments and cross-verifications, leading to strong conclusions in this
sensitive and immanently important topic. Several surprising and gratifying
results, which were registered in the robotics community to great interest, were
obtained.
In addition to the scientific output, the outcome of this thesis attracted also
significant public attention, confirming the importance of the topic for robotics
research.
The major parts and contributions of this thesis are described hereafter in more
detail. Furthermore, the resulting publications which are an outcome of the
work are cited."
5 comments
Comments sorted by top scores.
comment by Kawoomba · 2012-08-16T12:50:14.036Z · LW(p) · GW(p)
In this dissertation, the essential question about the necessary requirements for a safe robot is addressed in depth and from various perspectives.
That's right, we find out how to program robotic arm trajectories of industrial robots such that they don't hit humans, and also how to potentially make sure that robotic soccer players don't crush human soccer players when they fall onto each other.
Did I mention that this dissertation grants an actual Ph.D. in Computer Science from an actual reputed university?
the outcome of this thesis attracted also significant public attention
Glad that the whole Asimov safety issue finally gets more public attention. With "gratifying results", even!
For shame.
Replies from: Risto_Saarelma↑ comment by Risto_Saarelma · 2012-08-16T13:44:23.869Z · LW(p) · GW(p)
Isn't this just a solid piece of research about how to keep industrial machines from mangling humans? Safety engineering is a perfectly good topic to get an actual Ph.D. in, even if you don't address AGI friendliness.
Replies from: Kawoomba↑ comment by Kawoomba · 2012-08-16T13:55:46.562Z · LW(p) · GW(p)
Right. My issue is with framing the topic as "one of the most fundamental questions of Human-Robot Interaction: How to ensure safety in human-robot coexistence?", then naming the whole thesis for Asimov's first law, of all things. Then implying you made great strides in solving the "safety question". Asimov wasn't concerned with trajectory planning. Tying those together in the very title seems like grandiose equivocating to me.
From a cursory glance, I'd also object to the "solid piece of research", but let's not go down that particular rabbit hole.
Replies from: Risto_Saarelma↑ comment by Risto_Saarelma · 2012-08-16T14:05:28.572Z · LW(p) · GW(p)
I'm assuming the Asimov thing is a joke by the author who doesn't think of friendly AGI research as an actual thing separate from mainstream safety engineering or of robots as anything more than very narrow AI industrial machines whose interactions with people conclude with either an unfortunate mangling or a satisfactory non-mangling of the person.