Question about making turns

[jgeidl]

I've got the 3 wheeler as yet unnamed. When I make a turn with it, or
any I have ever built in the past, I drive one wheel forward and drive
the other in reverse. I got to thinking about it today and am
wondering if that is how most of you do it?

The alternative would be to brake one wheel and drive the wheel
opposite of the direction of the turn forward. In other words, for a
right hand turn I would brake the right whell and drive the left on
forward.

Is there an advantage in doing one over the other?

Thanks,

Jim, K6JMG

[Wayne C. Gramlich]

Jim:

The following paper could be useful:

[Butokim]

Wayne,

Did you, maybe, forget to attach the paper?

Jim, K6JMG

--
You received this message because you are subscribed to the Google Groups
"HomeBrew Robotics Club" group.
To post to this group, send email to hbrob...@googlegroups.com.
To unsubscribe from this group, send email to
hbrobotics+...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/hbrobotics?hl=en.

[Wayne C. Gramlich]

Jim:

(Sorry, 'bout that, I clicked the send button by accident.)

Here's the URL:

http://rossum.sourceforge.net/papers/DiffSteer/DiffSteer.html

The math near the end is a bit over the top.

In short, the robot will go in a left turn if Vleft < Vright,
a right turn if Vleft > Vright, and a straight line if
Vleft = Vright. The robot will turn in place if Vleft = -Vright.

Regards,

-Wayne

On 05/07/2012 05:17 PM, jgeidl wrote:

[anfederman@comcast]

If the third wheel is fixed, (not free to turn) it will act like a rudder,
and prevent the turn. Skid steering is also called tank style steering
(with tread it is real effective). For competition robots, a six wheel
(three per side) configuration works well, if you have the middle wheel a
little lower the the front and back. This of course is for flat terrain.

If you run the wheel is opposite directions, your turn about the center of
rotation - for two wheels, this will be in the middle of the axle. If you
power one side only, the robot the robot moves in an arc. How tight an
arc depends on gearing and such, but typically it is shallower then you
expect. If you power one side at 50% and the other at 100% a diff style
robot will only bank slightly. For accurate steering with a diff drive
system you really need wheel enocders.

If you are using FP motors, pushing the robot with the power off typically
generates enough back EMF that the fans and LEDs come on the PWMs.
--------------------------------------------------
From: "Wayne C. Gramlich" <wayne.gra...@gmail.com>
Sent: Monday, May 07, 2012 5:33 PM

To: <hbrob...@googlegroups.com>
Subject: Re: [HBRobotics] Question about making turns

[James Geidl]

Thanks guys.  I am drawing out the radii in AutoCAD to get a be perspective on how this guy will turn.  I guess there is nothing keeping me from doing it both ways depending on circumstance.  That's sort what I was asking; what situation makes one method better than another?

I am reading the document that that Wayne linked to.  So far it isn't saying anything special but does get into how to figure out will happen based on a LOT of calculations.  :-) 

Jim, K6JMG

To unsubscribe from this group, send email to hbrobotics+unsubscribe@googlegroups.com.

For more options, visit this group at http://groups.google.com/group/hbrobotics?hl=en.

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To post to this group, send email to hbrob...@googlegroups.com.

To unsubscribe from this group, send email to hbrobotics+unsubscribe@googlegroups.com.

[Wayne C. Gramlich]

Jim:

A differentially steered robot can traverse any arc. Even after
you crunch all the numbers in the paper, the numbers will probably
not do you much good, since the paper uses ideal motors, wheels,
H-bridges, zero inertia, etc. The real world has motor/gearbox
friction, changing battery voltages, surface inclinations, wheel
slippage, etc. I do not recommend over thinking this stuff.

Usually I run my robot with:

Left := Forward + Twist
Right := Forward - Twist

If Forward is 0, the robot turns in place and if Twist is 0, the
robot goes relatively "straight". If both are 0, the robot is
stopped. Care needs to be taken to ensure that Left or Right
does not exceed the 100% duty cycle on your H-bridge.

Regards,

-Wayne

On 05/07/2012 07:47 PM, James Geidl wrote:
> Thanks guys. I am drawing out the radii in AutoCAD to get a be
> perspective on how this guy will turn. I guess there is nothing keeping
> me from doing it both ways depending on circumstance. That's sort what
> I was asking; what situation makes one method better than another?
>
> I am reading the document that that Wayne linked to. So far it isn't
> saying anything special but does get into how to figure out will happen
> based on a LOT of calculations. :-)
>
> Jim, K6JMG
>
> On May 7, 2012 6:55 PM, "anfederman@comcast" <anfed...@comcast.net

> <mailto:anfed...@comcast.net>> wrote:
>
> If the third wheel is fixed, (not free to turn) it will act like a
> rudder, and prevent the turn. Skid steering is also called tank
> style steering (with tread it is real effective). For competition
> robots, a six wheel (three per side) configuration works well, if
> you have the middle wheel a little lower the the front and back.
> This of course is for flat terrain.
>
> If you run the wheel is opposite directions, your turn about the
> center of rotation - for two wheels, this will be in the middle of
> the axle. If you power one side only, the robot the robot moves in
> an arc. How tight an arc depends on gearing and such, but
> typically it is shallower then you expect. If you power one side at
> 50% and the other at 100% a diff style robot will only bank
> slightly. For accurate steering with a diff drive system you really
> need wheel enocders.
>
> If you are using FP motors, pushing the robot with the power off
> typically generates enough back EMF that the fans and LEDs come on
> the PWMs.

[snippage]

[Butokim]

Wayne,

Trust me, I don't not have the tools to over-think anything!

I'm not sure I understand your "Twist."

James M. Geidl, K6JMG
D.B. Cooper, you have a message.

-----Original Message-----
From: hbrob...@googlegroups.com [mailto:hbrob...@googlegroups.com] On
Behalf Of Wayne C. Gramlich

Sent: Monday, May 07, 2012 8:33 PM
To: hbrob...@googlegroups.com
Cc: Wayne C. Gramlich
Subject: Re: [HBRobotics] Question about making turns

--
You received this message because you are subscribed to the Google Groups
"HomeBrew Robotics Club" group.
To post to this group, send email to hbrob...@googlegroups.com.
To unsubscribe from this group, send email to

hbrobotics+...@googlegroups.com.

[Steven Nelson]

I used to teach a class on building drive trains for 1 lb. Ant weight Combat Robots. I made a simple inexpensive demo bot I called (Instant Ant). I covered a 3" wide by 4 inch long piece of plastic (Lexan) with self adhesive backed industrial Velcro.  I mounted some Dave Brown Lite flight wheels on two Servos modified for continuous rotation and covered the servo motors with the opposite Velcro that I used on the Lexan base plate. 

This way the servos could be easily stuck to the Base plate (instantly) in any configuration I wanted to demonstrate and you could compare how the motor placement affected steering and weight distribution once the battery and radio and other things were added and moved around as well. I found that this simple model taught me and others a lot about what happens when you move the motors, batteries/ heavy stuff and the wheels around. 

 I eventually built a larger one using that pre drilled angle iron and up to four motors and wheels which could be bolted in different configurations for comparing 4WD or 2WD skid steer. I used this model to develop the tread width to axle shaft center line ratio 1.34 to 1 I used when I built the 220 lb. Heavy weight robot Evelyn. When I first drove Evelyn I new exactly how it would handle and turn even though it was 14 times heavier than the small 15 lb model I tested.  

 I considered these two demonstration machines Real World Cad. They really taught me a lot of cool things about drive trains and weight distribution... I can usually just look at a 4WD robots lay out and tell how well it will turn. I've also measured a lot of my competitors machines tread width and axle shaft ratios to see how well their variations work in actual operations. There really is a sweet spot depending one the turning and drive ability performance your going for and simple small model can show you the way before committing to an expensive design and fun with the MIG welder or drill press...

I may just build another (Instant Ant) for the robot club I attend here in Sacramento because these questions come up often for every builder. Plug and Play engineering I kinda like that idea...;)

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To post to this group, send email to hbrob...@googlegroups.com.
To unsubscribe from this group, send email to hbrobotics+...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/hbrobotics?hl=en.

--
Inspection time: The period of time required to inspect a recently welded piece of metal using your bare hand.
The hotter the metal the shorter the period of inspection time

[Wayne C. Gramlich]

Jim:

Twist is the amount of turning. Positive Twist turns left and negative
Twist turns right. If Forward is positive the robot goes forward, and
if Forward is negative, the robot goes backward.

Regards,

-Wayne

[snippage]

[Butokim]

got it

Thanks,

[KM6VV]

What do you mean by tread width to axle shaft center line? Wheel base?
axle to axle for a 4WD?

Alan KM6VV

[Tim Craig]

Jim,

For differential drive robots, the turn radius is controlled by the
difference in wheel speed and the spacing between the wheels. Of
course, for wheels with finite thickness, this spacing may not, in
practice, end up being the distance between the centers of the contact
patches since there may be wheel irregularities and your shafts may not
be perfectly parallel to the ground. When planning a turn, you can
either drive one wheel faster, one wheel slower, or both. You can
always drive one wheel slower but if your robot is running fast, you may
not be able to increase the speed to get the radius you desire.

A problem that can show up in tank steer robots and possibly in 2 wheel
differential drive systems, if the third wheel provides enough drag, is
that the faster wheel(s) may over power the slower wheels and keep the
robot moving straight. In that case you may need to either brake the
slower side or even apply some reverse power to it to keep it from being
dragged.

As Alan pointed out, to do well controlled turns, you need some form of
feedback from the drives.

Tim

[anfederman@comcast]

Using teleop on a tank style robot is interesting. I find it is hard to do
with a single joytstick. It is a lot easier for a human operator to use two
joysticks, on for left one for right. That way one can get the robot to
steer straight by adjusting the joysticks.
It almost like balancing a broom stick on you finger.

If you just use two brushed DC motors mounted 180 from each other you run
into a problem, as the windings typically make the motor run forward faster
by about 10%. On GoPHR, the teleop program takes this into account, and
limits the maximum power to one of the motors.

--------------------------------------------------
From: "Tim Craig" <TimC...@Druai.com>
Sent: Tuesday, May 08, 2012 12:45 AM

To: <hbrob...@googlegroups.com>
Subject: Re: [HBRobotics] Question about making turns

> --
> You received this message because you are subscribed to the Google Groups
> "HomeBrew Robotics Club" group.
> To post to this group, send email to hbrob...@googlegroups.com.
> To unsubscribe from this group, send email to

> hbrobotics+...@googlegroups.com.

[Butokim]

Thanks, Tim. I realize that to do anything with these guys planned and
consistent I am going to have to deal with the odometry at some point. I
just want to get things move forward and backwards in a controlled fashion
before I take on the whole encoder issue.

James M. Geidl, K6JMG
D.B. Cooper, you have a message.



-----Original Message-----
From: hbrob...@googlegroups.com [mailto:hbrob...@googlegroups.com] On

Behalf Of Tim Craig
Sent: Tuesday, May 08, 2012 12:46 AM
To: hbrob...@googlegroups.com
Subject: Re: [HBRobotics] Question about making turns

[Butokim]

Thanks, Alan. The difference in forward and reverse speeds is something I
am experiencing. I am fiddling with the idea of running the forward wheel
slightly slowly the reverse wheel to see if I can compensate for this.
Thanks for bringing this up as I thought it was just and anomaly in the
hardware I have.

James M. Geidl, K6JMG
D.B. Cooper, you have a message.



-----Original Message-----
From: hbrob...@googlegroups.com [mailto:hbrob...@googlegroups.com] On

[Alan]

Having two sticks for steering a tracked vehicle started because of the
reduced complexities as opposed to a conventional steering wheel (Ackerman).
Each stick did clutch or brake, depending on which way it was pushed/pulled.
I don't think too many of our 'bots actually use a clutch, much less a brake
on each stick. And with our 'bots commonly having electronics between the
sticks and the actual control of motors, it is no big deal to drive with a
single two-axis joystick. Better, probably. The translation is not that
difficult. And how many people have caterpillar tractor experience anyway?

Maybe in a few years we'll be able to get inexpensive brushless motors
(already available for the R/C crowd) and then the directional brush effect
won't be a problem.

Alan KM6VV

> -----Original Message-----
> On Behalf Of anfederman@comcast

[anfederman@comcast]

My pleasure James.

Good judgment comes from experience. Experience comes from bad judgment.

Uneven rates are a problem even if you solved the problem mechanically by
running paired motors through gear boxes set up to account for f/r bias.
I've seen designs here the motors are side by side, one goes to the wheel
directly, the other has two equal spur gears and a long axle to reverse the
rotation. Not all motors/gearboxes/bearings are equal. Also brushed motors
may age differently. Differential drive is also called 'skid' steering.
Hard to do odometry control based on wheels not turning, or turning but
slipping and skidding!

It is interesting that the PR/2 and Texai do not use skid steering. they
use a drive unit that steers (Similar to a kids tricycle). I am guessing
there is a good reason for this - probably makes SLAM calculations a lot
easier.

Footnote: ( I believe FP motors had left and right hand gearboxes - though
those cars typically have conventional front wheel steering. Even if the
drive wheels are turning at different rates, the forward steering corrects
and drift.)
--------------------------------------------------

[Butokim]

Alan, You just brought up something else I have been thinking about, steering with my (presently) idler. Whole different subject unrelated to this but thanks for reminding me.

Unfortunately, this is where I get in trouble. In the midst of doing one thing I think about another and get distracted by it and never get the first thing finished. If there was big bucks for started yet unfinished projects I'd be rich instead of pretty.

Jim, K6JMG
--
Sent from my Android phone with K-9 Mail. Please excuse my brevity.

"anfederman@comcast" <anfed...@comcast.net> wrote:

My pleasure James.

Good judgment comes from experience.  Experience comes from bad judgment.

Uneven rates are a problem even if you solved the problem mechanically by 
running paired motors through gear boxes set up to account for f/r bias. 
I've seen designs here the motors are side by side,  one goes to the wheel 
directly, the other has two equal spur gears and a long axle to reverse the 
rotation. Not all motors/gearboxes/bearings are equal.  Also brushed motors 
may age differently. Differential drive is also called 'skid' steering. 
Hard to do odometry control based on wheels not turning, or turning but 
slipping and skidding!

It is interesting that the PR/2 and Texai do not use skid steering.  they 
use a drive unit that steers (Similar to a kids tricycle). I am guessing 
there is a good reason for this -  probably
makes SLAM calculations a lot 
easier.

Footnote: ( I believe FP motors had left and right hand gearboxes - though 
those cars typically have conventional front wheel steering. Even if the 
drive wheels are turning at different rates, the forward steering corrects 
and drift.)

[Maddox.Robert]

I’m an engineer, and I’m just getting into the whole robotics thing and planning on doing a Magellan style robot for maybe the next competition. Though I live in Reno I’d love to take a weekend to do some robotics stuff when life allows.

In any case, with the regard to the turning, it seems to me the missing point of either wheeled or track movement is a point of reference. Without an external point of reference all the accumulated slippages of wheels, gears, belts… and the inconsistency of the ground you’re travelling on, leads to an impossibility to match perfect world scenarios the mathematics presents.

It seems the only way to maintain an exactness in a moving system is to maintain an external point of reference, possibly using a camera which finds an object, and then after a move can measure a displacement from the previous measurement.

Now this whole line of thought might be overthinking a simple issue, but it seems correct to me.

[Tim Craig]

I believe the PR/2 setup gives them a holonomic platform, ie, the robot
can translate without rotation.

Tim

[but...@gmail.com]

I agree.  I really think that some kind.of external reference system is the only way to get around all the variables

Jim, K6JMG

"anfederman@comcast" <anfed...@comcast.net> wrote:

It is interesting that the PR/2 and Texai do not use skid steering.  they 
use a drive unit that steers (Similar to a kids tricycle). I am guessing 
there is a good reason for this -  probably

makes SLAM calculations a lot 
easier.

[Alan]

Hi Robert,

 

My son and I hope to compete in the next RoboMagellan as well.  Too bad you’re so far away!

 

Some general observations on the RoboMagellan “Navigation task”.

 

The camera is nice, but can’t typically give you exact location.  Stereo cameras and triangulation might get close. 

 

A Sony Kinect can give you quite a bit of information I’m told, and might let you resolve location to known fixed points on the course. 

 

Laser pointers and camera(s) or Kinect can probably allow you to triangulate your position, and possibly heading.

 

GPS can give you starting headings and distances (GPS limitations), and in-travel headings and rough locations.  Cameras or Kinect can help you identify targets and help you calculate headings and distances once they are within range (GPS gets you near).   They can also help identify obstacles and help you calculate headings and distance for avoidance. 

 

Odometry (wheel sensors or an odometer) is only partially useful, as varying terrain of the RoboMagellan course will contribute considerable errors.  It a help in detecting “stuck” wheels.

 

A Compass can help keep you on course, but it can be fooled by local or course metallic objects/buildings. 

 

Sonar/IR sensors can help you detect both goals and obstacles, but can’t do much to differentiate between the two.  Used in pairs or arrays, they can give you bearings to (or away) from a detected object.

 

Touch sensors are “last line of defense” or “final contact” devices, depending on the task.

 

A good multi-axis IMU can give you both a compass and inertial information about your progress (velocity and rotations).

 

All of these sensors need to be combined (sensor fusion) to give the best position information on your ‘bots location and heading.

 

Waypoints are used “plan the trip”.  Known obstacles and goals can be plotted on a map.  I believe we are given coordinates for the goals.

 

Alan

[Maddox.Robert]

I understand that- the reason I was at the last major competition/ show was my son who had Lego League entries. I saw the Magellan robots run and was fascinated by the complexity of what they were doing, then the thought that I could do it better. That’s probably without considering the enormity of the task, but that’s my usual driving force.

 

Now I’m thinking way outside the box- it is possible to construct a radar type of system, and then a rather complicated math program can in essence ‘graph’ the terrain, including the points of interest and any (and all) obstacles.

This would create a point of reference external to the moving body, yet still referenced from the moving body.

This would then see what’s going on from its position relative to all the other positions.

In essence what I’m suggesting is the view that the moving body is moving in relation to the environment can be turned around: the environment is changing and moving with respect to the body (which can be viewed as stationary). This is reminiscent of some older video games- like tank battle.

 

As for wheel stickage and sensors to detect that: if enough velocity is given to the vehicle the sticking becomes a non-issue. The issue then turns to a power source for speed- I imagine one of those antennae controlled off road truck models, but they’re usually powered by a liquid fuel- answer to that is several batteries and perhaps a 24 volt drive system.

 

From: hbrob...@googlegroups.com [mailto:hbrob...@googlegroups.com] On Behalf Of Alan

Sent: Tuesday, May 08, 2012 12:33 PM
To: hbrob...@googlegroups.com

[Austin Hendrix]

On 05/08/2012 11:12 AM, Maddox.Robert wrote:

I’m an engineer, and I’m just getting into the whole robotics thing and planning on doing a Magellan style robot for maybe the next competition. Though I live in Reno I’d love to take a weekend to do some robotics stuff when life allows.

In any case, with the regard to the turning, it seems to me the missing point of either wheeled or track movement is a point of reference. Without an external point of reference all the accumulated slippages of wheels, gears, belts… and the inconsistency of the ground you’re travelling on, leads to an impossibility to match perfect world scenarios the mathematics presents.

It seems the only way to maintain an exactness in a moving system is to maintain an external point of reference, possibly using a camera which finds an object, and then after a move can measure a displacement from the previous measurement.

Now this whole line of thought might be overthinking a simple issue, but it seems correct to me.

You're quickly getting into some pretty advanced territory there.

In general, all of the sensors on our robots are flawed in some way, and the best we can do is to use statistical methods to combine that data into a better representation of the world.

For position measurement, we generally use our odometry measurements as an initial guess at position, with some covariance, and then use some other sensor that detects local objects to figure where we are within that covariance. Odometry is critical in this process, since it limits the search space of possible positions. Several books have been written on the subject.

Traditionally, the sensor of choice for obstacle detection is a scanning laser rangefinder, since vision is a difficult problem by itself.

GPS is a good, globally-available reference system, but the accuracy is limited to several meters; if you want to know where you are to a finer degree, you still need some kind of localization. Various active radio and light-based triangulation systems have been created, and they can have better accuracy than GPS, but they require fixed transmitters which make them unsuitable for Robomagellan.

Accelerometers, gyroscopes and compasses can improve your odometry to some degree, which lowers the covariance on your position estimate.


If you're just getting started, my advice is to start with something simple and achievable, and make incremental upgrades from there. You'll be much happier if your make your mistakes one at a time, on a small robot, rather than all at once on a big project.

-Austin

[Alan]

I don’t know how many RADAR units you can get that size, but the Kinect can create depth-wise visual maps of what it sees, and the data can then determine a path through obstacles, or the proper path to advance to the goal and touch it.  You have a RADAR license?  ;>)

 

A set of appropriate “landmarks” could be sighted for bearing and range, and the updating position or “track” could be calculated.  Maybe a corner of a building, tree, or a walkway? 

 

I’m talking about getting “stuck”, where the wheels spin, no amount of velocity will get you out!  Behavior needs to “shift gears”, and try, for instance, backing up one way or the other.  Odometry (wheel encoders) would detect that situation.

 

Alan KM6VV

[Steven Nelson]

I measure tread width on a 4WD drive train from the outer edge of the right tire tread across the robot  to the outer edge of the left tire tread. I measure the axle center line from the center of the front axle shaft to the center of the rear axle shaft on just one side of the robot. Example Tread width measures 25 inches. Axle center line to axle center line to axle center line on the right side of the robot is 18. 5 inches that give me a tread width to axle shaft ratio of 1.35 to 1. I've also noticed that the black circle left on a hard surface (like asphalt) measure almost 18.5 inches after doing a hard full power skid steer turn with one side of the robot drive train in forward and the other side/ drive train in reverse.  So the distance between the center line of your axles has a great influence on the turning circle. If you increase this distance the robot will drive straighter but it will not turn as quickly. If you decrease this axle center line distance it will turn quicker. If you make this distance too short it will eventually become unstable and tend to rock fore and aft when you change direction like going from straight forward to straight reverse. It is also very important to try and mount all of your heavy components between the axles and as low as possible for stability. If you move a lot of weight past the distance between the axles it moves the load center and also causes the fore and aft rocking motion. So component placement the weigh of the components and the height above the axle all effect the robots stability. I was taking to one of the folks at Willow garage about this and he mentioned that on their telepresence bot that is actually 2WD that when they placed the monitor up high on the center pole they mount it on they had a initial testing surprise when the robot face planted the monitor into the ground. Sine we both come from the combat robot world we both laughed when I mentioned that they had invented a Thwackbot or at least a fine Hammer weapon. He told me they actually fixed the problem with software and a gyro to help maintain the balance. That Darn Newton stuff can be a bit annoying at times...;)

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To post to this group, send email to hbrob...@googlegroups.com.

To unsubscribe from this group, send email to hbrobotics+unsubscribe@googlegroups.com.

For more options, visit this group at http://groups.google.com/group/hbrobotics?hl=en.

[Alan]

OK, I see what you are referring to.  But I’d expect the skid marks from a turn-in-place to be more like 29.4”, the diagonal of the rectangle formed by the four contact patches of the tires.  However you don’t mention the width of a tire tread, so the center line of the tire tread could be used rather than the outside tread width.  That would bring down the diagonal length.  Not sure how the skid could just be the outside tread width, and not more like the wheelbase, unless you have TWO sets of skid marks.

 

But I like your empirical work!  Although I never made a study of the relationship, I do tend to favor a slightly longer four-wheel wheelbase to width.  I’ll measure my current 4WD ‘bot (made for a photographer friend) from SuperDroid parts when I get home tonight (he missed the last meeting)!

 

Yes, stability testing of carts or robots is necessary (even required for our medical robots).  I remember hearing the reports of the mechanical guys checking the required tip angles (I think they called them something else), and the angle of the inclines they were required to negotiate.   We have some pretty high vision and patient carts on the da Vinci.

 

Thanks the explanation!

 

Alan KM6VV

 

On Behalf Of Steven Nelson

[Alex Bravo]

It's a small world Robert. I used to live in Reno and work at IGT too. :)

If you want to understand how and why probabilistic localization works, I strongly suggest taking this class:

http://www.udacity.com/overview/Course/cs373/CourseRev/apr2012 

It definitely helped me to understand it.

Alex Bravo

[Maddox.Robert]

That looks like right up my alley- that may be what I’m moving toward. What do you with math, engineering and physics: go for programming a self-driving car.

That’s a technology which is based on experimenting with robots, or should be introduced into robotics.