Safe Robots
Title: "Singularity Clamping"
Here the LWRIII performs a reconfiguration from "ellbow up" to "ellbow
down" while exerting quasi-static forces on a human head. The head is
assumed to be clamped. This example shows the injuriy potential posed by
singularities in case of clamping. This is due to the fact that the robot
is theoretically able to exert very high forces which can easily lead to
various fractures. As a counterbalance a collision detection based on a
disturbance observer is used, leading to much lower injury potential.
Robotics: Science and Systems 2007:
Manipulation Workshop
Title: "Approaching Asimov's 1st Law II"
Here the influence of the robot mass and velocity during blunt
impacts with humans is shown. Several robots with weights ranging from
15-2500 kg at different impact velocities are impacted with a mechanical
human head mockup. This is used to measure the so-called Head Injury
Criterion, a measure for head injury caused by impacts.
The potential injury occurring during the actual impact will saturate
with increasing robot mass and is from a certain point on only depending
on the impact velocity. Furthermore, it will be confirmed that Severity
Indices focusing just on the moment of impact like the Head Injury
Criterion are not an appropriate measure of injury severity in robotics
because no robot exceeds their safety critical thresholds. This is due to
the usually significantly lower velocities of the robots compared to
impact tests carried out in automobile crash-testing.
From the presented impact tests, the rarely analyzed injury source
clamping is motivated by the breaking distance of the investigated robots.
Especially for larger robots clamping is an injury source one has to
focus on and for which countermeasures have to be found.
2007 ACM/IEEE International Conference on
Human-Robot Interaction
Title: "Approaching Asimov's 1st
Law"
The desired coexistence of robotic systems and humans in the same
physical domain, by sharing the same workspace and actually cooperating in
a physical manner, poses the very fundamental problem of ensuring safety
to the user and the robot. In order to quantify the potential danger
emanating from the DLR lightweight-robot (LWRIII), impact tests at the
Crash Test Center of the German Automobile Club ADAC were conducted and
evaluated.
A collision detection and reaction scheme, based on a
disturbance observer is used. It utilizes only the proprioceptive
capabilities of the robot and provides a filtered version of the external
torque.
The outcome of the dummy crash-tests indicated a very low
injury risk posed by rigid impacts with the DLR LWRIII. This was confirmed
by real human-robot impacts at robot velocities up to 2.5m/s. Based on
this experimental evaluation generalizations to robots of arbitrary mass
can be drawn.
Real Robot-Human Impacts
Title: "Physical Interaction and Impact Experiments"
In order to ultimatively prove that the impact experiments conducted at
the ADAC are reasonable, several human-robot impacts were carried out. The
impacted body parts were the head, chest, shoulder and abdomen.
The torque estimation is, apart from being used as a collision
detection mechanism, utilized as an adaptive scaling of time increments in
the trajectory generation and allows the user to intuitively push the
robot forth and back along its desired trajectory. Combined, these
mechanisms are used to distinguish between desired cooperation and
collision in physical human-robot interaction.
Initial Impact Experiments,Collision Detection & Reaction
Title: "Criteria and Control Structures for safe Human-Robot
Interaction"
- Impact experiment with the DLR light-weight robot III hitting the
outstreched arm of a human at different velocities
- Impact experiment with the DLR light-weight robot III hitting a
"dumm-dummy" of the human head-neck complex at different velocities
- Impact experiment with the DLR light-weight robot III hitting a
balloon at different velocities
When a collision is detected, several reaction strategies can be
activated sich as switching to gravity compensation mode or simply
stopping the robot.
our work is supported by and integrated in