Human Emulation - Summary
RobotServicesGroup.com already provides details about robots to compare their performance
including; price, power, speed and degrees of freedom. But what is missing is a measure of how
"good" are the robot's motions. With our Human Emulation project, we will attempt to standardize
the testing of robot motions for comparison against humans, and other robots.
RobotServicesGroup.com already provides details about robots to compare their performance
including; price, power, speed and degrees of freedom. But what is missing is a measure of how
"good" are the robot's motions. With our Human Emulation project, we will attempt to standardize
the testing of robot motions for comparison against humans, and other robots.
| All rights reserved. |
RobotServicesGroup.com
 | |  | |  | |  | |  | |  | |  | |  | |  | |  | |  | |  |
NOTE: These robots are not meant for children. All of the humanoid robot models discussed on this website use
powerful servos that can pinch or even crush your fingers if you are not careful. Consult the recommended minimum
age from the manufacturer before allowing children to use any of these robots.
powerful servos that can pinch or even crush your fingers if you are not careful. Consult the recommended minimum
age from the manufacturer before allowing children to use any of these robots.
Test Platform is level side to side
TEST DEVELOPMENT:
Forward Walking
We needed to develop scientific tests to objectively measure each motion. One test of a robot's
walking ability is to simply measure how long it takes the robot to walk a certain distance.
However, if we selected a set distance for each robot to walk (like 3 feet or 1 meter), it would
provide unfair advantage to robots with long legs, and disadvantage shorter robots.
Forward Walking
We needed to develop scientific tests to objectively measure each motion. One test of a robot's
walking ability is to simply measure how long it takes the robot to walk a certain distance.
However, if we selected a set distance for each robot to walk (like 3 feet or 1 meter), it would
provide unfair advantage to robots with long legs, and disadvantage shorter robots.
|
The second measure of walking skill is how close to a straight line the robot can walk. Distance is measured as
the absolute value of the variance from a straight line. For example, if the robot strayed 10 inches to the right on
the first test, and 10 inches to the left on the second test, then the average of the absolute value would be 10
inches, not zero. We will also divide the average variance from a straight line into the robot's height to get a
percentage. Doing this should keep the score comparable between robots of different sizes. For our testing we
did not permit any course corrections during a trial. If the robot veered to one side, we just kept on walking.
Another measure of walking skill is simply the number of steps it takes for the robot to complete the distance. In
order to complete the test, both feet must completely pass the finish line.
the absolute value of the variance from a straight line. For example, if the robot strayed 10 inches to the right on
the first test, and 10 inches to the left on the second test, then the average of the absolute value would be 10
inches, not zero. We will also divide the average variance from a straight line into the robot's height to get a
percentage. Doing this should keep the score comparable between robots of different sizes. For our testing we
did not permit any course corrections during a trial. If the robot veered to one side, we just kept on walking.
Another measure of walking skill is simply the number of steps it takes for the robot to complete the distance. In
order to complete the test, both feet must completely pass the finish line.
The I-Sobot is 6.5 inches tall and has to walk 19.5 inches.
The Robonova-1 is 12.2 inches tall and has to walk 36.6 inches.
The KHR-1 is 13.4 inches tall and has to walk 40.2 inches
The Robonova-1 is 12.2 inches tall and has to walk 36.6 inches.
The KHR-1 is 13.4 inches tall and has to walk 40.2 inches
| Variance from center is measured for each test |
Which robot more closely
matches Human motions?
matches Human motions?
Test Preparation
We tested the Robonova-1 from Hitec, I-Sobot from Tomy and the robot that started it all, the KHR-1 from Kondo.
Please note that there have been no modifications made to the robots whatsoever. They are the base model as
delivered directly from the manufacturer. The only exception is for the KHR-1 we added plastic feet and a remote
control, to make the tests consistent across all robots.
We performed each test on a vinyl floor surface on a test platform that we constructed ourselves. Other
preparations included leveling the floor surface, and setting the robot's home position as accurately as possible.
If the floor was not level, or if the robot's home position was off, then we would find the robot consistently
wandering to one side or the other.
We performed each walking test 20 times to build up an average result. For the 180 Degree Turn test, we did 10
turns to the right and 10 turns to the left to make sure we weren't favoring one leg. We also switched directions
after every two tests to make sure we didn't over-use any servos through repeated use. We took video of each
test, and digital photos of the finish location to measure for the straight line walking test.
Summary results are included below:
We tested the Robonova-1 from Hitec, I-Sobot from Tomy and the robot that started it all, the KHR-1 from Kondo.
Please note that there have been no modifications made to the robots whatsoever. They are the base model as
delivered directly from the manufacturer. The only exception is for the KHR-1 we added plastic feet and a remote
control, to make the tests consistent across all robots.
We performed each test on a vinyl floor surface on a test platform that we constructed ourselves. Other
preparations included leveling the floor surface, and setting the robot's home position as accurately as possible.
If the floor was not level, or if the robot's home position was off, then we would find the robot consistently
wandering to one side or the other.
We performed each walking test 20 times to build up an average result. For the 180 Degree Turn test, we did 10
turns to the right and 10 turns to the left to make sure we weren't favoring one leg. We also switched directions
after every two tests to make sure we didn't over-use any servos through repeated use. We took video of each
test, and digital photos of the finish location to measure for the straight line walking test.
Summary results are included below:
Humans can Turn 180 Degrees very quickly and easily,
but how well can robots mimic this motion?
but how well can robots mimic this motion?
180 Degree Turn
Most humans can change directions in a few
seconds, with only a few steps, but this task can
be challenging for robots. This is especially true
if the robot lacks servos in the hips or at the
waist. We studied the same types of data we
used to measure forward walking; length of time,
variance from a starting point, and number of
steps.
Most humans can change directions in a few
seconds, with only a few steps, but this task can
be challenging for robots. This is especially true
if the robot lacks servos in the hips or at the
waist. We studied the same types of data we
used to measure forward walking; length of time,
variance from a starting point, and number of
steps.
INTERPRETATION:
Forward Walking
The Robonova-1 walked three body heights in an average time of 50.5 seconds, the I-Sobot took 24.9, and the
KHR-1 took 17.6 seconds to cover the same relative distance, but a human only needs 4 seconds. While none of
the robots tested come close to matching human performance, what we can clearly see is that the KHR-1 is the
fastest robot we have tested. But the KHR-1 is the oldest robot, with a release date in 2004, how can this be?
The secret is in the walking motion provided by the manufacturer. For the Robonova-1, the standard walking
motion allows for 3 forward steps, then the robot stops and returns to the "home" position. In contrast, the I-Sobot
can perform continuous walking, merely by pushing the joystick on the remote forward and holding it. While the
KHR-1 does not have continuous walking, it is very close. The KHR-1 walks for 10 steps before returning to the
"home" position. Another contributing factor is that the KHR-1 has a longer stride than either the Robonova-1 or
the I-Sobot. This is proven by the fact that the KHR-1 only needed 24.7 steps to complete the test, and yet the
Robonova-1 and I-Sobot each needed more steps to complete the test (35 and 33.8). That means that each
stride on the KHR-1 is longer as a % of the robot's body height than the other robots.
While the I-Sobot had less variance in walking a straight line, the margin of victory over the KHR-1 is small. Keep
in mind that the I-Sobot includes a gyro as standard equipment, but the KHR-1 does not.
180 Degree Turn
The results were closer in the 180 Degree Turn test. Each robot completed the turn in roughly the same amount
of time (17.6 seconds vs 15 seconds). This came as a surprise, since the I-Sobot comes with a continuous
turning motion, but the Robonova-1 and KHR-1 do not. Where the KHR-1 really shined was in the variance from
center categories only needing 40% of its body height side-to-side and 9.9% of its body height front-to-back, as
well as the number of steps. This is an incredibly tight turning circle, with only twice as many steps needed as a
Human (wow!). And all this was achieved without the benefit of hip servos. Truly a great performance.
Forward Walking
The Robonova-1 walked three body heights in an average time of 50.5 seconds, the I-Sobot took 24.9, and the
KHR-1 took 17.6 seconds to cover the same relative distance, but a human only needs 4 seconds. While none of
the robots tested come close to matching human performance, what we can clearly see is that the KHR-1 is the
fastest robot we have tested. But the KHR-1 is the oldest robot, with a release date in 2004, how can this be?
The secret is in the walking motion provided by the manufacturer. For the Robonova-1, the standard walking
motion allows for 3 forward steps, then the robot stops and returns to the "home" position. In contrast, the I-Sobot
can perform continuous walking, merely by pushing the joystick on the remote forward and holding it. While the
KHR-1 does not have continuous walking, it is very close. The KHR-1 walks for 10 steps before returning to the
"home" position. Another contributing factor is that the KHR-1 has a longer stride than either the Robonova-1 or
the I-Sobot. This is proven by the fact that the KHR-1 only needed 24.7 steps to complete the test, and yet the
Robonova-1 and I-Sobot each needed more steps to complete the test (35 and 33.8). That means that each
stride on the KHR-1 is longer as a % of the robot's body height than the other robots.
While the I-Sobot had less variance in walking a straight line, the margin of victory over the KHR-1 is small. Keep
in mind that the I-Sobot includes a gyro as standard equipment, but the KHR-1 does not.
180 Degree Turn
The results were closer in the 180 Degree Turn test. Each robot completed the turn in roughly the same amount
of time (17.6 seconds vs 15 seconds). This came as a surprise, since the I-Sobot comes with a continuous
turning motion, but the Robonova-1 and KHR-1 do not. Where the KHR-1 really shined was in the variance from
center categories only needing 40% of its body height side-to-side and 9.9% of its body height front-to-back, as
well as the number of steps. This is an incredibly tight turning circle, with only twice as many steps needed as a
Human (wow!). And all this was achieved without the benefit of hip servos. Truly a great performance.
CONCLUSION:
The KHR-1 is our winner for Human Emulation. The KHR-1 was the fastest robot with the least variance from
center in our Forward Walking Test. The 180 Degree Turn Test was roughly a draw in terms of time, but the
KHR-1 had a tighter turning circle with fewer steps. The final deciding factor is that the KHR-1 was released more
than 4 years ago! This robot is the very first humanoid robot ever made for retail sale, and it still blows away
robots that were released much later. This is a clear indication that the folks at Kondo did a masterful job of
creating walking and turning motions that can hold up over time.
The I-Sobot was able to do a decent job of mimicking human motions because it came pre-loaded with continuous
walking motions, a gyro, and it is substantially cheaper then the other robots. But for our money, the best robot
for Human Emulation is the KHR-1 from Kondo.
Hopefully, as new robots are launched in the market they can build on these factors to keep improving
performance. Perhaps someday, there will be a robot that can walk and turn as quickly and as accurately as a
human.
The KHR-1 is our winner for Human Emulation. The KHR-1 was the fastest robot with the least variance from
center in our Forward Walking Test. The 180 Degree Turn Test was roughly a draw in terms of time, but the
KHR-1 had a tighter turning circle with fewer steps. The final deciding factor is that the KHR-1 was released more
than 4 years ago! This robot is the very first humanoid robot ever made for retail sale, and it still blows away
robots that were released much later. This is a clear indication that the folks at Kondo did a masterful job of
creating walking and turning motions that can hold up over time.
The I-Sobot was able to do a decent job of mimicking human motions because it came pre-loaded with continuous
walking motions, a gyro, and it is substantially cheaper then the other robots. But for our money, the best robot
for Human Emulation is the KHR-1 from Kondo.
Hopefully, as new robots are launched in the market they can build on these factors to keep improving
performance. Perhaps someday, there will be a robot that can walk and turn as quickly and as accurately as a
human.
| FREE Robot Test Drive Would you like to try out the I-Sobot or the Robonova-1 for yourself? Click here to sign up for a FREE Test Drive from the comfort of your own home. |
The KHR-1 is our current
Human Emulation champion
Human Emulation champion
To keep the test fair, we elected to set the distance equal to three times each robot's body height. So an I-Sobot (the
shortest robot in our Survey) will need to walk 19.5 inches, but a Manoi PF01 (the tallest robot in our survey) must
cover 47.1 inches. We can then compare the time for each robot to the time for an average human, hence, Human
Emulation. The time measure starts when the robot begins to move, and ends when both feet have crossed the finish
line marking 3 body heights.
shortest robot in our Survey) will need to walk 19.5 inches, but a Manoi PF01 (the tallest robot in our survey) must
cover 47.1 inches. We can then compare the time for each robot to the time for an average human, hence, Human
Emulation. The time measure starts when the robot begins to move, and ends when both feet have crossed the finish
line marking 3 body heights.
 |
| |
 |
| |
 |
| |
 |
| |
 |