New Dynamixel MX-28: info available on Robotis Support Site
New Dynamixel MX-28: info available on Robotis Support Site
by PedroR
Tue May 17, 2011 4:25 pm
Hi all
The Robotis Tech Support site has now got updated information on the actuator used in DARwIn-OP, the MX-28.
(This servo was formerly named RX-28M but Robotis moved it under a new series name.)
The page is here http://support.robotis.com/en/techsuppo ... /mx-28.htm
From the specs on the page, I have to say this is by far a huge leap forward in Dynamixel and is years ahead of the competition.
Highlights:
- Magnetic encoder with no blind zone (so you have resolution on the whole 360º range)
- Cortex M3 controller (a big improvement over the Atmega 8....)
- 0.08º resolution (if you do the math you'll see position is reported in the 0-4096 range)
- 1 wire TTL serial interface (it uses the same interface as AX-12 and AX-18 )
- Baud Rate up to 3 Mbps
- The page hints at true PID control and all the hardware involved:
"Feedback : Position, Temperature, Load, Input Voltage, etc"
I don't know if there is true measurement of voltage and load but if there is indeed true PID control, I guess these sensors (or accurate sensory estimation) is in place.
All the rest are the usual specs.
A big thumbs up to the increase of bus speed to 3 Mbps, the magnetic encoder which means precision won't degrade over time, the 360º position control (no blind zone) and quadruple the resolution (4096).
I'll wait and see about the TTL interface: it seems to have the same interface as AX-12 and AX-18 so you should be able to mix and match but you'd probably need to reduce speed to 1Mpbs.
Also at 3Mpbs I don't know how it behaves in terms of signal noise; probably the cables are improved...
Regards
Pedro
The Robotis Tech Support site has now got updated information on the actuator used in DARwIn-OP, the MX-28.
(This servo was formerly named RX-28M but Robotis moved it under a new series name.)
The page is here http://support.robotis.com/en/techsuppo ... /mx-28.htm
From the specs on the page, I have to say this is by far a huge leap forward in Dynamixel and is years ahead of the competition.
Highlights:
- Magnetic encoder with no blind zone (so you have resolution on the whole 360º range)
- Cortex M3 controller (a big improvement over the Atmega 8....)
- 0.08º resolution (if you do the math you'll see position is reported in the 0-4096 range)
- 1 wire TTL serial interface (it uses the same interface as AX-12 and AX-18 )
- Baud Rate up to 3 Mbps
- The page hints at true PID control and all the hardware involved:
"Feedback : Position, Temperature, Load, Input Voltage, etc"
I don't know if there is true measurement of voltage and load but if there is indeed true PID control, I guess these sensors (or accurate sensory estimation) is in place.
All the rest are the usual specs.
A big thumbs up to the increase of bus speed to 3 Mbps, the magnetic encoder which means precision won't degrade over time, the 360º position control (no blind zone) and quadruple the resolution (4096).
I'll wait and see about the TTL interface: it seems to have the same interface as AX-12 and AX-18 so you should be able to mix and match but you'd probably need to reduce speed to 1Mpbs.
Also at 3Mpbs I don't know how it behaves in terms of signal noise; probably the cables are improved...
Regards
Pedro
Re: New Dynamixel MX-28: info available on Robotis Support S
by Upgrayd
Tue May 17, 2011 11:35 pm
Very exciting information. Looking forward to the pricing information.
not exactly quadruple the resolution as the 1024 number space was over 300 degrees and the 4096 number space is over 360 degrees. So around 3.33 (10/3) times the resolution.
PedroR wrote: and quadruple the resolution (4096)
not exactly quadruple the resolution as the 1024 number space was over 300 degrees and the 4096 number space is over 360 degrees. So around 3.33 (10/3) times the resolution.
new dynamixel MX-28 <-> 01mech
by Stefan
Sun May 29, 2011 2:26 pm
I own a RX10, a few RX-24F and a RX64.
I always wished they would bring up a device with magnetic endless encoder.
Now I purchased a 01mech supermodified quarter scale servo.
It also has a 4096steps magnetic encoder and rs485 capability.
And it has an internal incremental counter for absolute positioning beyond 360 degrees.
I always wished they would bring up a device with magnetic endless encoder.
Now I purchased a 01mech supermodified quarter scale servo.
It also has a 4096steps magnetic encoder and rs485 capability.
And it has an internal incremental counter for absolute positioning beyond 360 degrees.
by PedroR
Wed Jun 01, 2011 3:38 pm
Hi Stephan
Thank you for your post.
We've been doing some math and research and we couldn't really figure out how you managed to get a much more affordable solution. Could you please explain as we'd be very interested in this as well?
The 01mech guys claim they usually work with Hitec Servos. A high torque Hitec servo (the MX-28 has 26kgf.cm I believe) is quite expensive and if you add the 45EUR of the cost of the board I couldn't quite figure out how it can be a lot cheaper.
Regards
Pedro
Thank you for your post.
We've been doing some math and research and we couldn't really figure out how you managed to get a much more affordable solution. Could you please explain as we'd be very interested in this as well?
The 01mech guys claim they usually work with Hitec Servos. A high torque Hitec servo (the MX-28 has 26kgf.cm I believe) is quite expensive and if you add the 45EUR of the cost of the board I couldn't quite figure out how it can be a lot cheaper.
Regards
Pedro
Teardown of the MX-28
by hizook
Mon Jun 20, 2011 3:52 am
Hey guys,
It looks like the MX-28 just became official. I was lucky enough to be a beta tester, and I wrote up some of my impressions (including photos of a complete disassembly!):
Review of the Robotis Dynamixel MX-28 Servo
I suppose y'all already know the major benefits: 4x resolution, 360-deg rotation, magnetic encoder. But I just got word... the MSRP is $219 USD. Too cool.
It looks like the MX-28 just became official. I was lucky enough to be a beta tester, and I wrote up some of my impressions (including photos of a complete disassembly!):
Review of the Robotis Dynamixel MX-28 Servo
I suppose y'all already know the major benefits: 4x resolution, 360-deg rotation, magnetic encoder. But I just got word... the MSRP is $219 USD. Too cool.
by PedroR
Fri Jun 24, 2011 3:11 pm
Hi i-Bot
To add to your comment we have 2 MX-28s which we Beta tested as well and indeed the Robotis software (ie Roboplus / Dynamixel Manager) does not seem to be fully ready yet to work with these servos at 3Mbps. (I am not sure if this limitation exists in the USB2Dynamixel physical hardware though as we've only tested with the Robotis software so far)
The servos came set at a factory default of 57.6kbps and you can change the speed setting using the Dynamixel Manager. However we could only access them using Robotis software at speeds up to 1 Mbps.
The setting to 3Mbps involves changing the Baudrate setting to value 252.
Theoretically this is the "divider" value (formula Baudrate(BPS) = 2000000 / (Data + 1) )and it is normally set to 1 for 1 Mbps operation.
What Robotis did here was add a special meaning to values 250-252 on the new MX servos meaning 2.25Mbps, 2.5Mbps and 3Mpbs. Dynamixel wizard still shows the "old" values (ie 252 = 7 905bps).
One other important point is that the processor on the servo is no longer a limited ATMEGA8 and is now a Cortex M3.
Proper PID control is also in place (you can tune all P, I and D parameters) and Load Calculation is now also done using PID (there is still no real current measurement as far as I can tell).
The Load calculation using PID seems to be much improved over AX-12 but we can't say it's perfect though.
Resolution is fair: it gives you the direction is which Load is being applied (CW or CCW) and from our tests we managed to get the load to report values up to 100 which isn't great but a good improvement over AX-12.
Responsiveness of the Load value parameters seems to have improved as well with the software reporting to Load changes relatively fast.
Finally with regards to your question about response time we have a Logic Analyser here but haven't hooked the servos up to it yet to verify the timings.
During the Actuated Character project the team found out that the AX-12s takes a random amount of time to respond to commands (the same command can get a response immediately or have random delays).
We haven't made this test on MX-28s yet though.
Pedro.
To add to your comment we have 2 MX-28s which we Beta tested as well and indeed the Robotis software (ie Roboplus / Dynamixel Manager) does not seem to be fully ready yet to work with these servos at 3Mbps. (I am not sure if this limitation exists in the USB2Dynamixel physical hardware though as we've only tested with the Robotis software so far)
The servos came set at a factory default of 57.6kbps and you can change the speed setting using the Dynamixel Manager. However we could only access them using Robotis software at speeds up to 1 Mbps.
The setting to 3Mbps involves changing the Baudrate setting to value 252.
Theoretically this is the "divider" value (formula Baudrate(BPS) = 2000000 / (Data + 1) )and it is normally set to 1 for 1 Mbps operation.
What Robotis did here was add a special meaning to values 250-252 on the new MX servos meaning 2.25Mbps, 2.5Mbps and 3Mpbs. Dynamixel wizard still shows the "old" values (ie 252 = 7 905bps).
One other important point is that the processor on the servo is no longer a limited ATMEGA8 and is now a Cortex M3.
Proper PID control is also in place (you can tune all P, I and D parameters) and Load Calculation is now also done using PID (there is still no real current measurement as far as I can tell).
The Load calculation using PID seems to be much improved over AX-12 but we can't say it's perfect though.
Resolution is fair: it gives you the direction is which Load is being applied (CW or CCW) and from our tests we managed to get the load to report values up to 100 which isn't great but a good improvement over AX-12.
Responsiveness of the Load value parameters seems to have improved as well with the software reporting to Load changes relatively fast.
Finally with regards to your question about response time we have a Logic Analyser here but haven't hooked the servos up to it yet to verify the timings.
During the Actuated Character project the team found out that the AX-12s takes a random amount of time to respond to commands (the same command can get a response immediately or have random delays).
We haven't made this test on MX-28s yet though.
Pedro.
by PedroR
Mon Jun 27, 2011 9:46 am
Hi billyzelsnack
With regards to the upcoming Robobuilder servos, I haven't heard any news from Robobuilder about this for quite a while.
From the exchange of emails with them almost 6 months ago I recall them mentioning true current measurement and they sent us a graph that showed quite accurate results.
However, we haven't had a chance to test them in real world so I can't really say if they're superior.
I'd say that the response time of the LOAD parameter and resolution is probably what would make the biggest difference and this is something that is heavily affected by how filtered the output is.
What I mean to say with all this is that without having tested the upcoming Robobuilder servos we can't really say.
It's also important to note that, although their brochure looks quite impressive this delay and lack of news from their end is a bit disappointing. This is not the first time they've released docs showing higher torque servos and later failed to release them (although I hope this will not be the case).
Back to the MX-28 what I can say, though, about the Load Feedback and the overall impressions about the servo is that LOAD feedback is significantly improved over other Robotis servos and seems now usable for real world apps. The resolution is not great but it works.
Also, the MX-28 feels like a very mature product and you get the feeling that it is a product that is significantly ahead of the competition.
Also, the fact that they chose to update the processor, implemented real PID control (which is tunable) and 3Mbps bus shows that they're working hard to push the edge on these relatively low cost servos (compared to those with planetary gears for example).
As i-Bot cleverly mentioned, one thing that is left to see if the efficiency of the bus protocol and if the servos respond in Hard Real Time (ie always take the same time to respond to each command).
Regards
Pedro.
With regards to the upcoming Robobuilder servos, I haven't heard any news from Robobuilder about this for quite a while.
From the exchange of emails with them almost 6 months ago I recall them mentioning true current measurement and they sent us a graph that showed quite accurate results.
However, we haven't had a chance to test them in real world so I can't really say if they're superior.
I'd say that the response time of the LOAD parameter and resolution is probably what would make the biggest difference and this is something that is heavily affected by how filtered the output is.
What I mean to say with all this is that without having tested the upcoming Robobuilder servos we can't really say.
It's also important to note that, although their brochure looks quite impressive this delay and lack of news from their end is a bit disappointing. This is not the first time they've released docs showing higher torque servos and later failed to release them (although I hope this will not be the case).
Back to the MX-28 what I can say, though, about the Load Feedback and the overall impressions about the servo is that LOAD feedback is significantly improved over other Robotis servos and seems now usable for real world apps. The resolution is not great but it works.
Also, the MX-28 feels like a very mature product and you get the feeling that it is a product that is significantly ahead of the competition.
Also, the fact that they chose to update the processor, implemented real PID control (which is tunable) and 3Mbps bus shows that they're working hard to push the edge on these relatively low cost servos (compared to those with planetary gears for example).
As i-Bot cleverly mentioned, one thing that is left to see if the efficiency of the bus protocol and if the servos respond in Hard Real Time (ie always take the same time to respond to each command).
Regards
Pedro.
by PedroR
Mon Jan 07, 2013 12:24 pm
Hi Unclebob
If you're only doing position based control (and not getting feedback from the servos to do closed loop control), you won't be taking full advantage of all that's new in the MX-28 but you'll see significant improvements on position control:
- Much better position control: the backlash on AX-12s tends ot increase with wear, This is due to both the wear on the plastic gears and also wear on the potentiometer, thus losing precision.
On the MX-28s this doesn't seem to happen (or at least they hold up much better). In addition, even though there is minor backlash on the MX-28, the fine position reading to the magnetic encoder seems to compensate this very well.
For example when DARwIn-Op is standing you can gently push forward or backwards; there's a small give but it will gently correct it.
If you go on to do closed loop control (for example reading back loads or detailed position), the MX-28 is vastly superior to the AX-12.
Reading back the load can be useful to detect, for example if you've hit an obstacle when moving.
As I posted before the load feedback mechanism is still not perfect but it's much improved and useful now.
For example if you want to build a gripper, you'll probably still need to use FSRs as there's not enough resolution on Load feedback if you want to control the force oft he grasping; using the servo feedback you'll probably be able to identify you've closed or grabbed something but in a very rough manner.
Another improvement on MX series servos is the Bulk Read command. This is a very important feature when you're doing true closed loop control (for self balancing for example) because the latency and time needed to query all servos is heavily reduced, compared to AX-12, thus getting much, much higher frame rates.
Regards
Pedro.
If you're only doing position based control (and not getting feedback from the servos to do closed loop control), you won't be taking full advantage of all that's new in the MX-28 but you'll see significant improvements on position control:
- Much better position control: the backlash on AX-12s tends ot increase with wear, This is due to both the wear on the plastic gears and also wear on the potentiometer, thus losing precision.
On the MX-28s this doesn't seem to happen (or at least they hold up much better). In addition, even though there is minor backlash on the MX-28, the fine position reading to the magnetic encoder seems to compensate this very well.
For example when DARwIn-Op is standing you can gently push forward or backwards; there's a small give but it will gently correct it.
If you go on to do closed loop control (for example reading back loads or detailed position), the MX-28 is vastly superior to the AX-12.
Reading back the load can be useful to detect, for example if you've hit an obstacle when moving.
As I posted before the load feedback mechanism is still not perfect but it's much improved and useful now.
For example if you want to build a gripper, you'll probably still need to use FSRs as there's not enough resolution on Load feedback if you want to control the force oft he grasping; using the servo feedback you'll probably be able to identify you've closed or grabbed something but in a very rough manner.
Another improvement on MX series servos is the Bulk Read command. This is a very important feature when you're doing true closed loop control (for self balancing for example) because the latency and time needed to query all servos is heavily reduced, compared to AX-12, thus getting much, much higher frame rates.
Regards
Pedro.
by UncleBob
Mon Jan 07, 2013 12:42 pm
Thanks Pedro.
I thought that backslash can be reduced by getting servos with more torque. Maybe I was wrong.
I was just wondering the real life differences between RX vs MX servos. The later have more resolution and communication speed. So I guess unless I need to do something really precise e.g. robot arm then I can get away with a cheaper one like the RX ?
I thought that backslash can be reduced by getting servos with more torque. Maybe I was wrong.
I was just wondering the real life differences between RX vs MX servos. The later have more resolution and communication speed. So I guess unless I need to do something really precise e.g. robot arm then I can get away with a cheaper one like the RX ?