PINNED - Roboteq Controller - developing for powerchairs

[Williamclark77]

The only specs that I can find on the BLT 500 is here... unfortunately there isn't a MilliOhms value.

I did try deriving it but the two methods I tried gave widely diverging answers.

http://www.goldenmotor.com/hubmotors/BL ... 0Curve.jpg

Beat me to it by a long shot. It took me over two hours to type my reply. I do have to at least pretend I'm doing something productive here

[Burgerman]

Start at 10. Go up in steps of 10. At some point, if Lenny's script is working, (and I didn't test this compensation yet) it will develop more torque at small stick movements. If you go even 10 more at this point it can get a bit wild and uncontrollable so beware!

Go test and find out! Probably best to have all delays set low/off as you test this. As it hides the affect.

[LROBBINS]

Here's an idea for dealing with poor motor current estimation at low currents. WARNING: I have no way of testing/tweaking this, so use at your own risk (and report the results please).

The problem Williamclark77 seems to be having is too little boost when starting out, yet too much as soon as he starts moving. I am guessing that at least part of this is because motor current is being underestimated at startup. It is possible, in the script, to have different compensation for low and higher current draws. Say, 1.5 times basic compensation if current is less than tot, and 1 if greater or equal to tot. Or, with a bit more arithmetic even to have a curved response; more compensation at low current tapering (linearly or following a curve) at higher currents, perhaps even zero compensation as you approach max amps (at max amps compensation is in any case effectively 0 as current can't be boosted anyway at that point).

I'm coming to like this idea despite the complexity, as it would be a way to allow strong compensation at low speed without risk of runaway, and this might be useful even if using accurate motor current sensors. BUT it will take a good bit of trial and error, and it would be helpful to measure and record motor current and boost amount and graph the relationship to get a picture of how these settings interact with feel. When I get to the point of writing a Roboteq script to go along with the CANbus controller I'll try to remember to include this possibility. As I do have current sensors on my prototype, I can also record Roboteq's estimated motor current and real motor current to get a picture of just how good or bad the estimated value is in wheelchair use. Can probably record the data on the SD card I'm using to store images for the display and log files, so I wouldn't have to lug a PC along (and wouldn't be distracted by watching the PC and driving at the same time).

Another possibility, though I don't know if it actually can be programmed, would be to have torque mode at very slow speeds (just for startup really) and transition to motor compensation mode as speed increases. Constant torque really is what one wants when turning from a dead stop - then the motors can't speed up after the casters start to swivel, but constant torque is absolutely not what one wants in order to move at a constant speed over varying terrain. I'll not work on this now, the plate's too full, but will keep it in mind.

Williamclark77 also mentioned something that I hadn't thought of, but think is actually useful. He's using different curving on the throttle and steering axes - in fact, a purely linear response for steering. In writing the CANbus master I included 4 levels, from none to strong, of exponential curving, but used the same parameter in both axes (as, I think, does Dynamic DX2). Williamclark77's setup is probably much better - a gentler response in throttle than in steering. So, last night I did a little re-write of CANbus master to have separate curving settings for the two axes, but haven't tested my code yet as I have things demounted from the breadboard in order to start packaging things in proper boxes.

Indeed, at low turn stick you might even want steering to be relatively stronger as one has to overcome lots of mechanical hysteresis to get the casters to turn and the drive tires to scrub.I'll leave the notion of stronger steering response at small joystick displacements for later, but if I decide to add that I'll just add selectable logarithmic curving so instead of a choice of 4 curves, there'll be say 7 - three with stronger response near center, then linear, then stronger response at big stick displacements. Won't add much to the code, but I will have to puzzle a bit to find a reasonable discrete-arithmetic way to mimic smooth logarithmic curves - doing real number arithmetic is just too, too slow (and not possible on the Roboteq at all - actually a good choice on their part, keeps people from complaining about non-responsive controllers if they've introduced the molasses of real number arithmetic).

Ciao,
Lenny

P.S. The Roboteq I will eventually buy will, unless new ones come along before then, be the HDC2450, the same one John is using. For it's physical layout, Rachi's present chair is stuck with its MicroMotor in-line brushed motors - very expensive, but compact and pretty efficient for PMDC geared motors. Don't really need anywhere near 150A capability, but that's the one Roboteq has with lots of IO pins and CAN capability (and the price difference from other models isn't all that great given the overall, growing, cost of this project).

[Burgerman]

The problem Williamclark77 seems to be having is too little boost when starting out, yet too much as soon as he starts moving.

No I suspect the opposite. He has it set to a low 10...

He tries to turn, or climb, and moves the stick a bit. Nothing happens. So he moves it further and further,

2 things happen... Acceleration curve catches up later. (needs increasing...)
It overcomes the resistance and then whoosh!

If I disable motor compensation on my PG Drives chair, same thing happens.

[LROBBINS]

Could well be. That's why I suggested setting MC to 0 as a test. If you're right this will make the behavior worse. On the other hand, having strong MC at low output tapering off as output increases does seem like something worth trying too. Could help a bit in the too-high MC full runaway case too, but only a bit. Much better to change the compensation parameter a little bit at a time and avoid that scary scenario.
Ciao,
Lenny

[Williamclark77]

I apologize if I wasn't clear. I have not done any testing with the script enabled. All of my seat time in it thus far has been using the Roborun software configuration tab only. I wanted to ask about the script before loading because it was written around BM's brushed setup. I wasn't sure if there would be a hardware conflict or known unsafe issue with mine being brushless. I just reread my initial reply about the script and see where I sound like I was referring to how it behaved with the script as is. My apologies.

It's 9:15 am here, 44 degrees F, windy, and rained all night, but is currently clear and not SUPPOSED to start back raining until tonight. I just put the chair on charge. A friend should be here in a few hours. I will try the script out and get her to video. That'll be a lot easier to show it's behavior than my words. And, not that I don't trust your coding abilities, but I would prefer someone be here in case I end up inverted or in the neighbor's pond

I will report back tonight on how it went and keep notes on my settings.

[Lord Chatterley]

Get an AB to move the joystick while standing beside [no one sitting in] the chair to test for caster-lag.

LC

[Burgerman]

VGet an AB to move the joystick while standing beside [no one sitting in] the chair to test for caster-lag.

???

[Burgerman]

I apologize if I wasn't clear. I have not done any testing with the script enabled. All of my seat time in it thus far has been using the Roborun software configuration tab only. I wanted to ask about the script before loading because it was written around BM's brushed setup. I wasn't sure if there would be a hardware conflict or known unsafe issue with mine being brushless. I just reread my initial reply about the script and see where I sound like I was referring to how it behaved with the script as is. My apologies.

How are you limiting reverse or turn speeds???

You wont get much success without the script.

Set reverse, and everything else to full speeds. Use the script to set up the user parameters. Leave compensation off.
Then after you get a sensible left/right rotation speed, and a sensible reverse speed set up, add motor compensation in small stages.

Set it up with motors unplugged initially, and use the roborun graphs to see what's happening...
Then with wheels off the deck and watch.
Then in a car park with an easy reach emergency pull OFF!

What are you using as an emergency stop solenoid in case of settings, programming, wiring error?

[LROBBINS]

Time for a warning!

The motor compensation part of the script has NOT YET BEEN TESTED by anyone. I don't have a Roboteq, and John didn't want to take time away from chair building, house remodeling etc. to hook up motors on the bench for testing. Until he enables motor compensation on his chair, we won't know if my algorithm works or just causes heaps o trouble. From his last comments, though, I see that John has tested the rest of the script, so at least the basic structure wasn't screwed up. Good, I've written a fair number of bugs in my years as an amateur programmer, but never wrote a program for which a bug could actually hurt someone. It's one thing to mis-calculate mitotic recombination frequency from number of spots on a fruit flies wings, it's another to have a friend driving something that relies on my calculations.

If it does work it should work for brushless as well as it just takes amps and milliohms to nudge the normal Roboteq channel setting by a bit less than the amount of back EMF.

Will has, though, set me straight on one thing. I was assuming that there was too much MC once the chair started moving, John thought that there is just too little all the time, and now Will says there's none - decidedly too little.

Ciao,
Lenny

[Burgerman]

Yes script works just fine on testing (with RC and me sat in it) and compensation set to zero.

Well I'm not dead, and it limits turn and reverse rates very well...

But get something set wrong in the roboteqs programming, a wiring issue, etc and you may well be dead... Under a truck. USE CARE! Examine everything for every eventuality, and the affect a failed wire or power supply, or whatever will have. Then fit something that you can hit that cuts battery power. Just in case...

[Burgerman]

Too little MC? Yes...
You don't move. So give it more, still don't move? Give it more again...
Once you overcome the inertia, bump, caster friction, ramp, then it will take off fast as you then have a large pulsewidth... The MC would have reduced this again fast. You cant do it that fast.

Worse, with turn rate at 20 percent, so your legs don't fly off at 50 volts, you may not get enough current until you are at "full stick" then whoosh!

MC also SLOWS the chair. With it removed it over runs way more...

[Williamclark77]

NOOOOW I get the warnings.... Too late!

In all seriousness though, I just survived 4+ hours in it. I tested the script quite a bit and tinkered with the settings. The MC does make a difference. Perfect? No. But there are other parameters: Input curves, acceleration/deceleration rates, amps, etc to test along with each different script change that was affecting how the MC performed. There should be a sweet spot in there. I recorded quite a bit of video and noted each change. I'm cutting my testing short for now. One of my dispatchers called in sick at work and I need to go cover their slot tonight. Nobody else would come in on a Friday night. If we are slow enough I will post my results tonight along with video.

I did go all the way up to 80 mohm on the MC setting. Does that sound too high? It was just starting to make a noticeable difference.

[woodygb]

I did go all the way up to 80 mohm on the MC setting. Does that sound too high? It was just starting to make a noticeable difference.

My estimations of the motors resistance derived from the limited info attached to the graph are in the 0.5 Ohm area.... your only just approaching 0.08 Ohm ...so I'd SUGGEST that your nowhere close yet.

http://openrov.com/forum/topics/brushle ... alculation


Find this ... Rm= motor resistance in Ohms...
Io= No load current.( see your graph pic )
V = Supply Volts 48v
Imax( see your graph pic )

re arrange this calc

CURRENT AT MAX EFFICIENCY:
Imax= sqrt(V Io/Rm)

[Burgerman]

On a 24v wheelchair, you generally need about 40 mOhm for powerful 4 pole motors. Going up to about 130 for weedy 2 pole motors.

The less power/higher impedance the motors the greater amount of compensation needed to give good control at small stick movements or over thresholds etc as you turn etc.

Set it too high and it gets very jerky. Set it a little higher still and it gets dangerous and leaps about uncontrolled...

[Lord Chatterley]

VGet an AB to move the joystick while standing beside [no one sitting in] the chair to test for caster-lag.

???

William said - 'Apply a little more and TORNADO! Both situation can be avoided by cleverly choosing my routes or turning at slight angles, but I know it can be done better.'

To which BM replied - 'Start low, and you will gradually get more "power" at small stick movements. Add a bit at a time and test. OUTDOORS. So it should turn without needing a big handful. DONT add too much at once. You will regret that!'

I meant it can be tricky getting the weight distribution right in a newly designed chair. Programming may be OK - problem could be stiff casters on a nose-heavy chair. Eliminate the simple stuff first.

LC

[LROBBINS]

The worst of the few wheelchair motors I have personal experience with had an internal+wiring resistance of 300 mOhm. There are actually some pretty good 4-pole motors with as much as 100-110 mOhm resistance. Ciao, Lenny

[woodygb]

Imax 11.563 Amps @ Max Efficiency
Io 1.282 No load Amps
V 48 volts
Rm ?
η max 84.22% Max Efficiency

Using
Imax= sqrt(V Io/Rm)
Rm = 0.460244 Ohms

BUT

Using
η max= (1- Sqrt ( Io Rm / V ))²
Rm = 0.253 Ohms

[Burgerman]

I meant it can be tricky getting the weight distribution right in a newly designed chair. Programming may be OK - problem could be stiff casters on a nose-heavy chair. Eliminate the simple stuff first.

LC

Yes. But they would need to be quite bad to have any affect on a properly programmed chair. I have no idea where mine point. And don't much care. It goes where I tell it and the casters do whatever caster do to follow suit.

[Williamclark77]

Thank you very much gentlemen. MC was making a little difference, but not enough to really matter from 0 to 70. It could've also been the placebo effect - meaning I was expecting and looking for results. Anyway, at the highest mohm that I tried today, 80, it was becoming noticeable. I will push it higher when I get back home tomorrow and video. Woody, thank you for taking the time to do those calculations. These motors are not regarded as the top-of-the-line by the electric biking community, but their requirements are much different than ours. I'm impressed with them so far though.

How are you limiting reverse or turn speeds???

Common sense before today. I will wire in a potentiometer soon. But as far as globally limiting it, nothing but restraint with my go finger.

As of a few hours ago the script is limiting reverse/turn - albeit far too much for my tastes in the default configuration. Not in total amount, but in responsiveness. Sans script - While going full speed and lightly turn right a few degrees, the chair stayed full speed and made a quick dart to the direction it was aimed. With the script - performing the same maneuver, the chair quickly loses forward speed and slowly starts moving toward where it is aimed. It seems as if all forward power is cut and only the steering motor pulling. I did not have time to alter the turn scaling though. I will as soon as time allows and give it a go. To turn off reverse and turn scaling, should 0 or 100 be input into their respective places?

I have run into something that concerns me with the Roboteq software. I have the upper and lower limits of both joystick inputs set as emergency stops. The steering input, motor 2, works as expected. Disconnect the joystick wire (or reach either limit I set, regardless of what it is) and it shuts the controller down. Throttle, which is motor 1, does not. Disconnect the wire and it goes wide open. I've tried configuring it pretty much every possible way. Still goes wide open as if "0" is input by the joystick. I've tried resetting to default, updating the firmware, and loading different profiles I had saved from my old hbl2360 controller. The Roboteq is simply not acknowledging any max/min setting on that channel. I will upload screen shots of my configuration when I get home if needed. I will read up on the manual also and see if mixed steering changes something.

Here is the video of my testing the low joystick input on it today. I tried not to put my hand over the joystick so that its movement could be seen. Apparently the audio didn't make the trip through cyberspace. Figures. I'll reupload it later. Ignore the blue wire. That's the usb for the Roboteq. Too hard for me to reach it and plug it in, so I let it ride along. Note: I'm letting off of the joystick as soon as it starts to turn in the first few clips with no scripting to prevent it from spinning.

Click it.

[Burgerman]

My comments. Not meant to be insulting or anything but think carefully. Try to understand what I am telling you. Because you have several severe problems.

You have lots of power with that 13s 45v battery and that roboteq controller. WAY more than any mobility controller. I think you have my battery?, and the same controller power as me. About 13,500 watts! That's 18+ horsepower... On tarmac it will spit you out and smoke its tyres then probably run over you. It may LOOK like a powerchair, it certainly wont perform like one!


1. Your biggest problem. You do not have remotely adequate hand / finger control or your hand on the joystick pod side for adequate stability to be driving a chair that powerful. Shoving a tiny joystick roughly about with part of your hand in approx. the right direction simply wont work. At very least, looking at the vid, you are going to need some of those very time delays etc that turn a typical accurate linear joystick into slow action slam it anywhere and wait pudding stirrer that I hate so much and that all powerchairs have until I remove it all...

I don't know how you will ever get over that physical / technique problem. Its like trying to do micro surgery with boxing gloves on.

2. You have a totally unworkable, massively too higher turn rate. No-one could drive that. I couldn't. Expo certainly wont help there! Reduce its turn speed drastically. Same with reverse. Start with 10 percent of what you have. So full stick moves it slowly. Or not at all. You want full stick left to rotate it once every 4 seconds at its fastest seen from above when the chair is empty and on a smooth surface. To start with. If it wont turn, add MORE compensation. Set motor acceleration and deceleration in roboteq software so it takes about .6 or more to reach full stick movement. Its way too fast. Reduce amps, forward speeds etc for now.

3. You also seem to have the same brushless motor problem that stopped me using them. No starting torque. They don't like starting up. Maybe not, hard to tell when 1 and 2 make it so uncontrollable.

4. Motor compensation will make it better at small stick movements, but you have turn rates and speeds set so high I cant see how you can possibly tell what its doing, especially due to 1. above... Since you cant get small movements and have no "feel" or accuracy.

I drove myself about via RC quite easily. With no compensation set at all. Reasonably smoothly on guessed settings. I don't remember what they were but turn and reverse were set to very low figures.

Speed looks good. How fast is that?

[Burgerman]

As of a few hours ago the script is limiting reverse/turn - albeit far too much for my tastes in the default configuration. Not in total amount, but in responsiveness

THIS is your SETTINGS problem.
Responsiveness comes from adding motor compensation. NOT setting turn or reverse rate too high... Set them low and slow. Add lots compensation till it responds! Repost results.

Script sets max speeds, turn speeds etc. absolutely as it should.

MC sets responsiveness. If it will not turn, add more MC! And it should then turn. Concentrate ONLY on slow reverse or slow turn. If not then it doesn't work.

[Williamclark77]

Joystick control is no issue at all. I was INTENTIONALLY pushing it that way so my hand wasn't in the way of seeing what they joystick was doing. I was holding the camera over it with one hand, but then you couldn't see the tires well. This was the best solution I had. Not to mention it was 44 degrees and drizzling rain. I was wet and freezing!

The turn rate is too high IF you push the joystick far enough. Yes, I've already said the max turn rate is too fast. The acceleration rate of the turning is not though. Hence why it's been fine for me so far, for testing only, to get the rest of the chair sorted. For a permanent solution it will be slowed. The max rate it will go that is, but not the acceleration rate.

In closed loop mode starting torque is plenty, even zero turning or on a steep hill. Steep enough my c350 will not go up without a rolling start, so I know it is possible. In open loop starting is fine in every situation except on a steep hill and zero turn only.

The video was to show nothing but the zero turn (like tank steering) issue. Nothing else. I wasn't trying to show where it works well. Those parts don't need fixing.

[Burgerman]

Read the post above as I was writing while you were. That IS your issue then. Repost result!

The only way to find the correct level of MC is to add too much in small stages and it gets to jumpy. And this will feel worse when the stick is too sensitive with too high turn rate. Then you have a feel for it. So you are easily able to go back a small amount to where it works best.

Send me a copy of your roboteq settings and script settings to burgerman@ntlworld.com so I can see what you set up?

So how fast is it? Looks about as fast as mine I think and faster than you expected?

[Williamclark77]

Yes, I agree with you. What you are saying is what I mean to get as the end result. I know what it needs to be. It's finding out what combo of settings will do it. I haven't gotten that far yet. My above post refers to how it's programmed as of now. Nowhere near correct and a bit worse than straight without scripting, but the ability to get it there is available now - it's finding that magic combo of settings that will take me a while. Hopefully less time with yaw's input.

My setup has never been attempted before and I have no baselines to start from. It's 14s and 3p Headway 15ah cells. Mathematically it will do 13.5 mph at 1000 motor rpm. Realistically, I think 12.5 mph. Top speed is not my main goal though.

[Burgerman]

I watched that video 8 times now. And have been thinking.

Your slowing while steer input at speed is caused by the tank steer algo you have chosen. It gives the best speed, but a reduced speed input to allow steering when you steer. Choose the one that leaves headroom and it wont happen. But you will lose about 2mph.

See the roborun screen while selecting the script, then graph the motors output and the control inputs. And choose A B or C tank mix then wiggle the stick and watch...

The locked wheel on a stationary turn, does not happen with my brushed motors. Its smooth. I suspect its something to do with the brushless motors starting up at very slow speeds where they don't seem to like it. What happens if you try to drive at say under 1 meter a minute? And I hate to think so, but its why I didn't try them. With the groove motors it turns smoothly. And will drive really really slowly or turn on the spot seamlessly. You may end up lowering gearing or having to go brushed? (At least by Radio control mines smooth. No analog joystick connected yet!)

[Williamclark77]

Without the script it doesn't slow while turning at speed. It will build enough speed in a 15 foot radius circle to flip (ask my test pilot/guinea pig/brother). It's midnight here. I'm going home. I will work on getting MC acceptable tomorrow and report back. I think once i get it right the rest will be easy(er) . Thanks to you, Woody, and Lenny.

[Williamclark77]

Without the script it doesn't slow while turning at speed. It will build enough speed in a 15 foot radius circle to flip (ask my test pilot/guinea pig/brother). It's midnight here. I'm going home. I will work on getting MC acceptable tomorrow and report back. I think once i get it right the rest will be easy(er) . Thanks to you, Woody, and Lenny.

[Burgerman]

isn't that because without the script its possible to turn at up to 100 percent speed? that's a 30mph difference in wheel speed? That ought to do something!

[LROBBINS]

Trying to think through this "slow down while turning" bit with the script, but not without. The script doesn't change the mixing algorithm of the Roboteq, so I don't really understand this UNLESS it's an interaction with motor compensation at full throttle. That is, as you approach full throttle and then add the resistance of turning (or at least starting a turn) or if you are still accelerating, MC will boost power, but only up to a point - 100% PWM. At that point, as one motor is slowed down in order to turn and the other can't be sped up any more, the chair would slow down. So, is this what's happening? The simple test is to see what happens with the script but with MC set at 0. If this thinking is correct, the behavior should then be the same as with no script. (A permanent cure would then be to change the MC calculation so that it tapers off to zero at say 90% of full current. However, there's no use re-writing the program without hard data about what's happening.)

User settings in the script, in general, are percentages of max. So, 0 is no output, 100 is full output. For example, setting steering scaling to 20 would make full L-R deflection equivalent to 20% forward stick or 20% of (already scaled) aft stick.

I'm very much with John (Burgerman) about how to do the testing. Set everything low and slow to start with. It's safer, you'll have more time to sense what's happening, perceptions will be less confused by still accelerating while steering etc. It will let you work on responsiveness and control, the first order of business. Using the speed pot, or reducing the numerical value for the "speed pot not present" equivalent, turn things down to a low value. That reduces everything proportionately without changing the relationship between fore-aft-steer-MC and so on. Once you've got responsiveness where you want it at low power, you can increase speeds in small increments and see what fine tuning is needed. You have a LOT of power at your beck and call and that's exciting, but you have to be in command and I don't think you'll be able to adjust control issues while doing wheelies in the mud.

BTW-the Roboteq brushless controllers do not have 150A output, so Will's controller can't sent as much power to the motors as John's. The motors, however, should be somewhat more efficient and the chain drive probably causes less loss than a worm drive. In any case, it's obvious from the video that there's oodles of power there.

Ciao,
Lenny