PDA

View Full Version : Reverse Knee and Humanoid Mechanics



lnxfergy
10-06-2009, 12:40 PM
I had the bot walking without the guns, but not 100% stable. After adding the guns, it wouldn't take more than 1-2 steps before falling over. My advise, don't bother with the Reverse Knee design unless you get into the high end dynamixel servos. 5990's have enough power to do a Reverse Knee, but, when you add all the extra weight, they don't perform very well.. I think I'm going to remove the scout bottom torso, re-work him into a standard huminoid.

Thanks, Connor

I wouldn't say that "reverse knee" is the problem -- it's the extreme knee angle, anything that far from lined up vertically is going to put a lot of force on your servos.

-Fergs

[MOD NOTE: This thread was created by splitting off posts from Connor's Shadow Scout Thread (http://forums.trossenrobotics.com/showthread.php?t=3003) lnxfergy was replying to a post there, and I've preserved that quote to try to leave some continuity. Fergs didn't [necessarily] create this thread, he was just "lucky" enough to be the earliest post in the split. The thread was split at the request of the OP for the original thread - best I can do unless someone else has a better idea... --- Adrenalynn]

SystemDefect
10-06-2009, 01:11 PM
I Hope the reverse knee setup isn't the problem as I've just spent my robot fund on servos and brackets for it.

DresnerRobotics
10-06-2009, 01:44 PM
Reverse knee is fine; it doesn't really matter which way the knee bends, it's mostly just aesthetics. What does matter is how the legs are aligned when weight is placed on them, and this is the problem that most people building 'reverse knee walkers' run into. Think about it, walking around with your knees bent is harder than walking with them mostly straight.

Mechwarrior and other mecha genres have polluted our minds with reverse knee walkers. Sure it looks pretty cool for them to have a huge angle in the knee bend, but it's far from efficient. The only bonus to it is that it lowers your COG. Ideally, you want all of your weight bearing servos to be aligned somewhat like this:

http://forums.trossenrobotics.com/gallery/files/1/4/9/2/legalign.jpg

But even with this design I sacrificed some payload for aesthetics... but I had some room to do so.

Connor
10-06-2009, 05:09 PM
When I refer to the Reverse knee design, I'm talking about the way the Scout was originally designed with the bent knee's. Tyberius is right on in the fact that the scout doesn't line up the servos correctly, which put a extreme amount of strain on the servos. When I tried it out with the servos all lined up and all the weight bearing down on the brackets and not so much on the servos themselves the servos hardly had any issues.

Thanks, Connor

SystemDefect
10-06-2009, 05:35 PM
I have my heart set on a reverse knee design but with the size of the payload I'm going to carry I need to build something with more torque then a Scout.

sam
10-06-2009, 06:39 PM
Thanks for the reply Connor!

Would adding a spring of some sort from the back of the reverse knee to the back of the hip help?

Or maybe some sort of shock absorbers like Tyb had on the Johnny 5?

Still, those must had unnecessary weight for the biped.

something like this :

http://forums.trossenrobotics.com/gallery/files/1/5/4/9/scout1_thumb.jpg (http://forums.trossenrobotics.com/gallery/showimage.php?i=1846&c=3)

Sam

Connor
10-06-2009, 08:41 PM
Thanks for the reply Connor!

Would adding a spring of some sort from the back of the reverse knee to the back of the hip help?

Or maybe some sort of shock absorbers like Tyb had on the Johnny 5?

Still, those must had unnecessary weight for the biped.

something like this :

http://forums.trossenrobotics.com/gallery/files/1/5/4/9/scout1_thumb.jpg (http://forums.trossenrobotics.com/gallery/showimage.php?i=1846&c=3)

Sam

I tried that, didn't work too well.. caused major issues with the walking sequence because the tension on the spring isn't linear. The deal is, they're just too much stress on the servos.. it needs to stand more upright. Also, the torso on the scout was too far forward, the weight needs to be OVER the HIPS, not in front of them.. I used the battery as a counter weight, but, that still wasn't enough. I'm afraid the only way to do it would be stronger servos or, with a more normal walking gate and upright stance and get away from the crouching stance. Move batteries to the feet.. and then hopefully your good to go.

Thanks, Connor

sam
10-06-2009, 09:15 PM
Ah OK,

Too bad because a crouch robot is awesome. :( I will keep what you said noted for when I make my bot.

Thanks for all the information Connor:)

(400th post! yeah!)

SystemDefect
10-07-2009, 08:19 AM
I want to Avoid having the batteries in the feet. That is just more weight you must lift with every step. I know it would help keep your COG very low but I think the answer is transferring the weight from one leg to another. My main battery pack will be located on the hips.

Orac
10-07-2009, 10:22 AM
I agree, batteries in the torso would make for a better and more natural walk if you can keep it stable. Will you be using gyros in the torso?

lnxfergy
10-07-2009, 10:30 AM
The question inevitably becomes: what portion of the overall weight is batteries? I'd guess that for most mechs, the battery weight is actually a fairly small portion of the overall weight: guns, servos, structural components and control mechanisms are the majority of your weight by far.

In such a case, batteries in feet can actually be useful, since otherwise you would end up with WAY too much weight up top, and little changes in position can cause huge changes in COG, meaning that the inaccuracy of the servos could tip the bot over by overshifting the COG.

-Fergs

SystemDefect
10-07-2009, 11:36 AM
Very good Point, I'm going to do my best to distribute the weight evenly but I'm not sure how successful I will be ;)

But in any case, I will make a cool looking Mech, so when I loose I can at least loose with Style ;)

Adrenalynn
10-07-2009, 12:20 PM
I do a lot of hiking, and it seems to me like whether I strap the backpack to my leg or wear it on my back - I still have to lift the same amount of weight with each step. The distribution of the weight over the balance point is another topic - but the weight isn't going anywhere.

DresnerRobotics
10-07-2009, 01:51 PM
I do a lot of hiking, and it seems to me like whether I strap the backpack to my leg or wear it on my back - I still have to lift the same amount of weight with each step. The distribution of the weight over the balance point is another topic - but the weight isn't going anywhere.

Weight distribution between robots and humans isn't really a valid comparison, given that we use a walking gait that very few robots are actually capable of even trying to mimic (closest robots have is ZMP (http://en.wikipedia.org/wiki/Zero_Moment_Point)). As humans, we think: "Weight on feet is bad!" because our natural bodies are used to walking with a high COG, and we essentially walk in a controlled fall forward (See ZMP, which is more or less what is trying to be mimicked there), and our bodies naturally adjust for added weight. If we walked long enough with weights on our feet, our bodies would condition themselves and we'd end up compensating for that as well.

With robots, there is no pre-conditioning. Evenly distribute the weight to the best that you can, and it will help your robot's overall balance, and thus help it walk better. And yes, that certainly allows for batteries in the feet. Giger wouldn't be able to walk without the batteries in his feet, because he's be far too top heavy, again, its all about even weight distribution.

Adrenalynn
10-07-2009, 01:55 PM
I completely agree with all your points. I was just observing that the weight doesn't transfer to the place Odo's extra mass goes just because you moved it up. Moving weight down is a no-brainer for getting the COG lower. Think more like cars than humans. The weight didn't increase just because I moved it lower, but it's a lot less likely to flip over, burst into flames, and ruin my afternoon... ;) This is why Sumo wrestlers squat.

SystemDefect
10-07-2009, 03:22 PM
To dispute lynn, I also do a lot of hiking and I'm not strapping backpacks to my legs. By shifting weight to the leg that is not being lifted from the ground you reduce stress on the servos. Having excess weight located in the feet would only cause extra strain.

SystemDefect
10-07-2009, 03:27 PM
I Agree with Tyb, About Robots and humans, but robots are adapt at precision and repetitive tasks.

SK.
10-07-2009, 04:50 PM
I do a lot of hiking, and it seems to me like whether I strap the backpack to my leg or wear it on my back - I still have to lift the same amount of weight with each step. The distribution of the weight over the balance point is another topic - but the weight isn't going anywhere.
Its of course correct that the weight stays the same, but its not only statics but dynamics that matters here. Weight attached to a body part moving more or less uniformly (like the upper body, moving forward with speed v while walking/ running) will have a dramatically different effect on energy and power requirements than the same weight e.g. attached to the feet, where its inertia has to be overcome to cycle its velocity between 0 and 2*v with every step.

Adrenalynn
10-07-2009, 05:01 PM
I'm gonna let it go... We'll leave it at "I disagree". You're assuming an absolute stiff upper body and absolute stiff leg.

Build it any way you wanna build it. We have fire extinguishers...

SK.
10-07-2009, 05:10 PM
Reverse knee is fine; it doesn't really matter which way the knee bends, it's mostly just aesthetics. What does matter is how the legs are aligned when weight is placed on them, and this is the problem that most people building 'reverse knee walkers' run into. Think about it, walking around with your knees bent is harder than walking with them mostly straight.
Of course, but most biped robots are in fact using a bent knee gait that looks a bit like a human needing to go to the bathroom to some extend (Bioloid, RoboNova, Omnizero, RoboCup robots..).
While Im not a specialist in walking robots I suspect that the kinematic singularity with unbent knees is the culprit here. The more you straighten the leg while walking, the less height change of the foot you get for changing the related servo positions. So Id guess there has to be some sweet spot for every design that weights foot agility vs. stress on the knee servo.

DresnerRobotics
10-07-2009, 06:39 PM
Of course, but most biped robots are in fact using a bent knee gait that looks a bit like a human needing to go to the bathroom to some extend (Bioloid, RoboNova, Omnizero, RoboCup robots..).
While Im not a specialist in walking robots I suspect that the kinematic singularity with unbent knees is the culprit here. The more you straighten the leg while walking, the less height change of the foot you get for changing the related servo positions. So Id guess there has to be some sweet spot for every design that weights foot agility vs. stress on the knee servo.


Yeah there is a definite balance and I believe we're on the same page here. I have both types of gaits on Giger currently; one which is very human-like, long stride, and maintains a mostly upright posture (however I'm having issues maintaining stability), and one which has the knees slightly bent, short-quick stride, the body leaning forward, and is much more stable for me. The latter of which benefits from lower COG and more lift range on the legs, at the cost of more stress on the knee servos. The previous really needs complete IMU feedback and ZMP calculation to pull off effectively.

I'm not really suggesting that all humanoid robots walk with stiff, straightened legs; what I'm saying is that if you look at the angle of a lot of 'reverse knee' walkers, that angle is pretty extreme even in comparison to the standard hobby humanoids you mentioned. It isn't ideal at all.

BDP
10-10-2009, 07:50 AM
Hi all, I found this site after seeing the very impressive 'hagetaka' biped while looking around online. I've always had an interest in robotics and in particular bipeds. Looking at the load problems with the crouching reverse knee biped, one system came to mind that I thought I'd ask your opinion on. My assumption so far is that it is not used due to inaccuracies in the mechanics playing havoc with the coding. I would like to hear any opinions however on where this design falls over.

Excuse the very shoddy image! If you hacked a servo or made your own hybrid by having the motor/gearbox (in green) drive a lead screw and the feedback pot measuring the angle at the pivot (in red), this should reduce the strain on the knee joint servo and allow the knee to be bent giving the correct 'look' or allow a greater payload for a straightened leg. The lead nut would need some application of an anti backlash method to prevent play from becoming too much a problem and would also need to pivot. I guess the leg would still only be as good as the weakest servo but even if this method does not take much load off the servo while it is driving the leg, I would imagine it should help reduce the servo work load while holding a position, in a way reducing the duty cycle.
http://img205.imageshack.us/img205/4271/leg.jpg

Adrenalynn
10-10-2009, 11:18 AM
Welcome to the forum!

As drawn - there has to be a bearing block out by the motor. That motor can't hang in space like that unless you're wearing safety glasses [and body armor for the larger examples...] for when that screw shatters. ;) They're also incredibly slow where servos are very very fast. If you have a look at the formulas for lead screws you can see why.

Assuming you've worked through both of these already - then I'm not understanding from your drawing - would you explain further?

BDP
10-10-2009, 11:27 PM
Thanks Adrenalynn! I agree the motor cannot just hang there, my intention was to illustrate the idea as basically as I could because pictures in MS paint tend to look more and more unclear as I add detail... perhaps that is just me though :veryhappy:.

I'm not at any level where I could build a biped, but looking at the reverse knee that idea came to mind and I thought I'd put it out there and see why no one seems to use it. Slow speed seems as good an answer as any, even if some shaft torque was sacrificed for speed at the gearbox it may turn out to be no stronger than a decent servo and would likely be still too slow.

JonHylands
10-21-2009, 07:42 PM
In terms of using a lead screw/linear actuator, you can build amazing humanoids with them. Probably not to the scale of a typical servo-based humanoid.

Look at MIT's M2: http://www.ai.mit.edu/projects/leglab/robots/m2/m2.html

It uses series elastic actuators, which are a variant of what you're describing (see Yobotics (http://yobotics.com/)).

I plan to someday build a reverse-knee velociraptor using series elastic actuators.

- Jon

kdwyer
10-24-2009, 11:30 AM
Worm-gear drives are extremely resistant to backlash, ie. it will hold position without using force (electricity.) The biggest problem with worm gears is speed. So, if you can beat that one...

JonHylands
10-24-2009, 12:01 PM
Well, speed is doable, as long as you have sufficient power. Brushless motors have tremendous power. If you can get a motor running at, say, 6000 rpm, and you're using 3/8-16 threaded rods, that means you have to turn 18 rotations to move the actuator one inch. 6000 rpm is 100 revs per second, so you would get almost 6" per second in that scenario.

- Jon

sam
10-24-2009, 01:04 PM
Worm-gear drives are extremely resistant to backlash, ie. it will hold position without using force (electricity.) The biggest problem with worm gears is speed. So, if you can beat that one...

Speed doesn't have to be an issue. It's all about the gearing.

I have a BLDC motor next to me that does about 12000 RPM theoretically no load. Put that on linear actuator and it can go really fast with less pushing force or really slow with lots of pushing force.

EDIT: The problem with most actuators like this is that worm drives are often much less efficient than other gears.

Adrenalynn
10-24-2009, 01:57 PM
The biggest issue isn't just "speed" it's efficiency.

And don't forget that there is a limit to how fast you can spin lead screws, which can be calculated based on all the other variables.

DC Robotic
10-25-2009, 10:24 AM
@ kdwyer, I thing you mean "backdrive". Backlash is the "slop" or "play" in the mechanism. Wormgears are resistant to being backdriven, but often have some backlash.

Along with Adrenalynn's efficiency concern. A lot of that lost efficiency in a ballscrew or wormdrive is manifested in heat. When you run mechanical devices that have sliding contact interfaces like screws as fast as you are talking about they get real damn hot real fast, not to mention wear.