PDA

View Full Version : Advantage of Dynamixel actuators?



Fatfrank69
02-10-2013, 02:03 PM
Hi Guys,

I'm currently looking into building a full size humanoid robot (yeah, I know...) to start with it will be teleoperated but eventually I will want to implement some autonomous capabilities using ROS possibly or another control system I am currently working on. I've noticed a lot of people are using Dynamixel servos for humanoid robots and robot arms on larger robots etc. What are the advantages of using these over regular servos with a similar torque?

Is it purely because of the feedback capabilities of the Dynamixel? When it comes to implementing some autonomous behaviour I don't see how the robot can interact with the environment (ie pick up an object) without feedback telling it where its arm is? Or are there ways you can implement this with regular servos?

tician
02-10-2013, 02:41 PM
High torque, high voltage (AX at 12Vmax; MX at ~14.8Vmax; RX/EX at ~18.5 Vmax), high accuracy, and simple UART control with position and some "load"/PWM feedback with the really expensive versions (MX-64/MX-106/EX-106) having actual current sensing. The dynamixel pro look really nice, but have not heard anything about them or the new micro servo in some time.

Hobby servos use PWM duty cycle input to set the goal position with most offering no position or load feedback. Most are also designed for 5~7.4 volts maximum. There are replacement control boards like the openservo, but still voltage limited and rely on the original hardware of the servo.


At a full size humanoid, you are looking at MX-64 as absolute minimum if not MX-106. At last glance, Charli-L from RoMeLa used a PLM design using EX/MX-106 and spring counterbalancing (keeps the legs mostly straight until the servos try to bend) while the Charli-H version used a combination of linear actuators a bit like a human/terminator design.

Gertlex
02-10-2013, 02:41 PM
For me, the quality of engineering is a big factor, as well. The servos tend to be high quality, and the servo cases are much better for robustly mounting than the standard hobby servos' cases are. The daisy chained wiring between servos also mean fewer wires, which is a definite nice thing to have.

Edit: And as Tician mentioned, higher voltages. 6 V is inconvenient to do with LiPos.

jwatte
02-10-2013, 06:05 PM
What are the advantages of using these over regular servos with a similar torque?

First of all, I don't know of a hobby servo that has torque/performance similar to the MX-64.

You can get a Hitec robot servo for about $120 (HSR-5990TG) that has 30 kgcm torque at 7.4V. This servo has a hole for an idler in the back. It still uses PWM control, but you can connect using a HiTec servo programmer to set some parameters of the servo. I haven't seen any indication that you can actually read back feedback at runtime. It also still does not do a full 360 degrees of movement -- it does 180 degrees only.

For $300, the MX-64T has 74 kgcm of torque at 14.4V, and supports true current sensing. It also uses a real absolute position encoder with 4096 counts per revolution, which is generally more accurate than most potentiometer-based position sensors.

At $45, you have the AX-12. It uses a potentiometer for about 280 degrees of position sensing, but can actually rotate all 360 degrees IIRC. At 16 V, it only has 16.5 kgcm torque, so it's weaker than the HiTec, but it's also half the price, and still supports the serial bus / TTL communication with feedback.

Dynamixel actually has nothing good at the $120 price point. The closest robot servo I know of is the Herculex DRS-0102. You might want to buy one of those and compare for yourself.

tician
02-10-2013, 08:34 PM
16V would kill an AX-12, so guessing you meant 12V. The AX-18A are capable of a bit more torque, but the older AX-18F had some overheating problems in the knees of a Premium humanoid. I still haven't gotten around to building humanoid legs out of the four bulk boxes of AX-18A yet to compare. Just haven't had enough users to warrant it - we need more students!!! The robots want YOU to join the UGA College of Engineering! Seriously! Three DARwIn-OP, the electronics for a fourth, an omni-wheeled rover, multiple kinects and other cameras, multiple Type-A humanoids, and plenty of extra parts to build whatever you want. That's just one lab in engineering, and there are several other labs in the engineering, computer science, and artificial intelligence departments dealing with robots and augmented reality. Sorry for the off-topic rant.


Another thing I didn't really detail: MX-series have a 12-bit absolute magnetic encoder connected to an ARM Cortex-M3 via SPI, while AX/DX/RX-series have a potentiometer connected to the 10-bit ADC of an ATmega88. All are capable of continuous rotation, but the potentiometer has a 60~80 degree dead zone and will wear out with use. IIRC, the herculex servos use ATmega88 and potentiometers just like the AX/DX/RX dynamixels although they seem to be a bit more feature rich.

Really wish I knew more about the price point of both the dynamixel pro and the micro servo prototype they were showing off on the CM-900 forum (pan-tilt controlled by CM-900 and an android phone for face-tracking). It is pretty annoying that there is such a large price gap between the AX-18 and MX-28, especially with the marginal performance difference between the MX-28 and MX-64 (and the somewhat frequent overheating/motor death we've experienced with the MX-28).

jwatte
02-11-2013, 01:06 AM
guessing you meant 12V

Yes I did. I guess I can't read :-) http://www.trossenrobotics.com/dynamixel-ax-12-robot-actuator.aspx

The AX-18 have always seemed like a strange beast to me. They are way up there on the speed scale, but not much better than AX-12 in torque. I bet a half-speed, double-torque version would suit most builders much better. But I guess it would compete too much with the MX-28 or perhaps RX-24 then?


there is such a large price gap between the AX-18 and MX-28, especially with the marginal performance difference between the MX-28 and MX-64

I think you mean that the AX-18 (or even AX-12) and MX-28 are not different enough in raw specs to warrant a 3x (or even 5x) difference in price. The difference in performance between MX-28 and MX-64 is pretty big, and the cost difference is only about 25% -- if you're going for MX-28, you might as well go for MX-64 IMO...

The MX-28 at $150 would be a well priced knock-out offering.

So, you want students, ey? Do you have full scholarships that cover a family of five? :-)

tician
02-11-2013, 07:48 AM
I think you mean that the AX-18 (or even AX-12) and MX-28 are not different enough in raw specs to warrant a 3x (or even 5x) difference in price. The difference in performance between MX-28 and MX-64 is pretty big, and the cost difference is only about 25% -- if you're going for MX-28, you might as well go for MX-64 IMO...
That is exactly what I was intending to state, but my brain dropped the ball, yet again.


So, you want students, ey? Do you have full scholarships that cover a family of five? :-) Depends on how many grants your major professor has gotten recently. The biggest disincentive for me is that it is in georgia, the place where few things make sense. Also, the tp/gop gerrymandered us really badly, so we still have a rep who thinks evolution and modern cosmology are "lies straight from the pit of hell. (http://religion.blogs.cnn.com/2012/10/10/congressman-draws-fire-for-calling-evolution-big-bang-lies-from-the-pit-of-hell/)" ಠ_ಠ

sarendt
02-11-2013, 08:28 AM
Really wish I knew more about the price point of both the dynamixel pro and the micro servo prototype they were showing off on the CM-900 forum (pan-tilt controlled by CM-900 and an android phone for face-tracking). It is pretty annoying that there is such a large price gap between the AX-18 and MX-28, especially with the marginal performance difference between the MX-28 and MX-64 (and the somewhat frequent overheating/motor death we've experienced with the MX-28).

Tician - You mention overheating/motor death with the MX-28; the MX-28 is used in DarwinOP correct? Is that the bot you have seen the motor's overheating and deing in? Any more thoughts or ideas on why this is happening? Also, do you think a improved bipead design would solve or reduce the problem? Have you seen or thought of any improvements in the vain?



The biggest disincentive for me is that it is in georgia, the place where few things make sense. Also, the tp/gop gerrymandered us really badly, so we still have a rep who thinks evolution and modern cosmology are "lies straight from the pit of hell. (http://religion.blogs.cnn.com/2012/10/10/congressman-draws-fire-for-calling-evolution-big-bang-lies-from-the-pit-of-hell/)" ಠ_ಠ

Distance degree's in EE available?

Thanks!
scott

tician
02-11-2013, 09:34 AM
The DARwIn-OP uses MX-28 to great effect. There have been others on the robotis forum reporting similar heat death of multiple MX-28 servos, but it does seem mostly preventable. The best way to prevent it is to allow the servo to completely cool down to room temperature after an overheating shutdown (just press the reset button of the CM-730 and it cuts all power to servos). If you don't completely disconnect power and let it cool, then repeatedly pushing it will cause it to reach some tippping point. Once past that point, it will seize up when it cools off and likely burn out the H-bridge when next powered on. If you notice the joint is frozen, don't test it; immediately disconnect it and begin the RMA process. It takes only seconds to burn out the mosfets if the motor is seized and only fractions of a second if the windings have shorted out. Several times the servo worked fine before putting the bot in the box overnight only to find a frozen joint in the morning.

Somewhat related are the gearsets: they appear to have changed material and/or surface treatment of the output gear/spline and that extra hundredth of a millimeter makes the bearings a definite press/interference fit instead of the hand press/slip fit of the older gears. I had to modify a hobby pinion puller to completely press the magnet side bearing onto the shaft of the last replacement gearset. Given the press fit and minimal clearance behind the magnet-side bearing, there is no chance of removing and reusing these bearings (as if that were ever an acceptable procedure) when the next output spline shears off or main gear gets a damaged tooth. I really hope they start shipping the replacement gears with new bearings already pressed on, or at least have new bearings included in the kit. Three so far with the output spline sheared off, and two others with a damaged tooth in the output gear causing it to stick a bit in one part of the horn's rotation - all from falling off a four foot tall stage.


I think the ME and EE degrees are starting up this fall, but I don't think they will be doing distance yet. I know GA Tech did distance MS degrees for working professionals to beef up their resumes, and I expect they still do.

sarendt
02-11-2013, 10:15 AM
I think the ME and EE degrees are starting up this fall, but I don't think they will be doing distance yet. I know GA Tech did distance MS degrees for working professionals to beef up their resumes, and I expect they still do.

Working professional... thats me - more or less

Cheers,
Scott

Fatfrank69
02-16-2013, 02:54 PM
Thanks for your replies. There are actually other high torque servos out there I forget the name of a company beginning with I that makes mega torque servos, plus servo city do a 4,000 oz-in one.

Charli from Romela is exactly what I'm going for. In the latest version they have removed the springs as they were having trouble writing algorithms to take these into account. In some joints there are two servos, if I assume they are the max torque that Dynamixel do (EX-106) that means he has around 200kg-cm torque in that joint. What do you consider to be the minimum torque my robot will need in the legs? Which joints are the ones requiring the high torque?

My current thinking is that I have the following options-

- Hobby servos (not strong enough)
- Linear actuators (too expensive)
- Servo city gearboxes (seems the best option now this could give me up to 244 kg-cm)
- Custom DC motor with servo control board (see below)

I'm also considering some cheap servos such as the notorious MG995 (whether or not I add open servo I don't know). I hear they are very jittery, would this still be a problem in a servo gearbox?

What do you guys make of these?

http://www.ebay.com/itm/170916157858?ssPageName=STRK:MEWAX:IT&_trksid=p3984.m1423.l2649

If I were to go for non- dynamixel servos what would I be missing out on? In particular in terms of later using something like ROS to autonomise the robot. Is it still possible to create an arm that can see and then pick up an object using ROS without dynamixels? Every robot I have seen so far uses dynamixels to do this.

jwatte
02-16-2013, 09:42 PM
Every robot I have seen so far uses dynamixels to do this.

Other people are as interested in saving money as you are. This quote should tell you something...

It is *possible* to do anything, but once you've compensated for all the weaknesses, you've ended up spending as much money, and a lot of time.