Demonstrate Danger, Safely
source link: https://hackaday.com/2022/07/30/demonstrate-danger-safely/
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.
Demonstrate Danger, Safely
Wouldn’t adding the capability to limit the maximum speed and torque of the robot arm greatly improve the safety?
Talking about robot safety in this instance is basically victim blaming – saying the kids should have known how to play chess with an industrial machine.
Did it never occur to anyone not to put high powered robots in a room with a bunch of kids? Someone needs to lose their job. Especially since solutions are so simple – anything from soft rubber grippers (at the least) to non industrial (3D printed) robot arms of which I’ve seen dozens here.
Also, sensors. The one thing everyone here at HaD is incredibly good at is attaching sensors to processors. I can think of at half a dozen that could work for this situation.
Yes, the robotic arm weren’t particularly fit for the application.
At the very least it should have a suitably torque limited grip and feedback for how large of an object it has gripped. If too big, then something else beyond the chess piece itself is in the grip, therefore it should release.
Limiting overall movement speed is also often a wise consideration around humans. But even here feedback is nice to have. If the power needed to move suddenly increases beyond expectations, then one has hit something and it is therefore wise to stop. (and even a lot of industrial robots do this. But this is more to protect the machine itself, since the speed of movement is often sufficient to be dangerous, especially when accounting for the time it takes the machine to stop moving.)
But likewise, sometimes these sensors providing our feedback fails in themselves. So it can be wise to have redundancy here, and if a disagreement is noted between our redundant sensors then we likewise stop. (there is far too many machines that fails spectacularly due to a broken sensor and lack of redundancy… Boeing’s angle of attack sensor is a good case study here.)
Limiting overall power of a machine to a level that is deemed “inherently safe” for the application is also something to consider. All though not always an option.
However, safety is often far from trivial in practice. There is often a lot of nuances one don’t notice before hand.
But likewise, a lot of safety issues can be avoided if development is given some time to focus on safety, instead of rushing towards product-release/project-completion.
-
Foldi-One says:
It may well have had those, I can’t find a better detailed breakdown of the accident – but a kids finger is likely smaller than the chess piece, and it doesn’t take much force at all to break a finger, so even tuned down to a power level so weedy it can barely hold/move its own weight…
-
Alexander Wikström says:
Fingers are rather fragile things indeed.
However, the machine should have a rather decent idea of where a chess piece is.
And therefor know when to grasp, and also know what diameter it is looking for.And even if a child’s finger is smaller than the chess piece doesn’t make our assessment if we picked it or not all that different.
We expect grasping something of X diameter, if we see something that is Y = X +/- 1 mm, then yes it is likely the chess piece, if it is wildly different, then it isn’t what we expected, therefor we should release.
(however, non round pieces will be a thing to consider differently.)
Another safety feature is to have a load cell for the head of the arm, not directly measuring grip force, but rather the tugging force of whatever it is holding. If the thing even shows a bit of force, then it likely isn’t the light weight chess piece, but rather something heavier, or something connected to something else. And therefor we should release.
Then there is also the idea of using an optical fence. If hands comes into the courtyard, then we stop. Just IR diodes and receivers is simple enough, and shouldn’t have major issues indoors.
-
The Commenter Formerly Known As Ren says:
Or a metal piece on top of each chess piece and an electromagnet inside of a gripper on the robotic arm.
-
Alexander Wikström says:
Yes, a magnetic mechanism for lifting the pieces would be a lot lot safer for the application.
-
Foldi-One says:
Yes it knows where the chess piece is, doesn’t mean it knows a kid that presumably hasn’t been told better has managed to put their finger in the path (and the operator didn’t react and hit the e-stop early enough either) – even a magnet gripper wouldn’t always help here. Anymore than a infant trying to body block an adult is going to work out well for that infant – the power and mass behind even a really lightweight mechanical arm can be way more than substantial enough to break the as we have mentioned rather delicate fingers…
-
Davidp says:
A visual system that can see the target could limit the possibility of a finger getting clamped. A Pixycam could identify/confirm the piece being moved, but all you’d really need is enough processing to say “chess pieces have round tops–Hey that (finger) isn’t round” and stop the movement until the path is clear.
-
Foldi-One says:
The only ‘safe’ solution is for the robot to only make its moves while the human is holding two buttons down that are far enough away and apart to be sure the human can’t physically be in the danger radius. And even that isn’t really certain, and I’d argue would actually make the robot on the whole more dangerous to be around in this case – its still potentially lethal but now everyone assumes its safe even when its got power as the permission to move buttons are not depressed.
But software can glitch and buttons can stick or outright fail short/open without it being apparent to the user. So if everyone is treating it like it can’t hurt them no matter what…
-
Dan says:
A large guillotine I saw once used not just two push buttons, but they had metal buttons and ran a small continuity test through the person touching them to ensure that they had two hands of the same person on it, not something accidentally leaning on a button or something.
-
Alexander Wikström says:
” the power and mass behind even a really lightweight mechanical arm can be way more than substantial enough to break the as we have mentioned rather delicate fingers…”
If the arm moves sufficiently slow (0.1 m/s or less), it wouldn’t be much different than placing a hand onto an almost stationary object. Even if it has practically infinite power behind its movement, it won’t have sufficient speed to rapidly impart any energy into whatever it is hitting. (and it isn’t like children grasp a lot of other moving objects on a daily basis)
One can also cushion the blow by padding the arm, but then one is likely moving too fast.
However, it wasn’t speed that broke the child’s finger.
The action that broke the 7 year old’s finger were when the arm placed the chess piece back down on top of it.Here a simple solution is to have a spring loaded gripper able to retract 1-2 cm (with feedback, could just be the E-stop), this would give the machine sufficient time to notice that things aren’t as they should. Since the spring loaded part started retracting ahead of expected and therefor the program knows that it has hit something that shouldn’t be there. Likewise does the spring loaded mechanism give the person more time to pull their finger away (since it wouldn’t apply much clamping force to speak off), something the child tried. But since the machine were far too rigid, they simply couldn’t.
-
Foldi-One says:
>If the arm moves sufficiently slow (0.1 m/s or less)
If it moves sufficiently slowly you die of old age waiting for it…
Not that you are wrong, but it pretty much entirely negates the point if it can’t function in a sane timescale. And even at 0.1m/s if you go and put your hand between it and the destination it will still crush it with ease (unless as well as being very slow it has no mass and/or stiffness).
-
Alexander Wikström says:
“And even at 0.1m/s if you go and put your hand between it and the destination it will still crush it with ease”
I see someone completely forgot the main point I have been talking about this whole time.
So I will reiterate again…
One shall have feedback for how much power is currently used for a given point in time during a move, and that one shall have an anticipation of when an increase in force is expected and not.
If one moves from X to Y and knows one expects to hit nothing along the route, then one don’t expect much force at any point along the route. If force is encountered, then something is wrong and one shall stop. (some inertial loading one will have to account for if one desires higher acceleration.)
If one moves from X to Y and knows that Y will be at a position one expects force, then it is a question of tolerances and latency for how much before Y we consider an increased force as expected. But still, if the force increases too early, then something is yet again wrong and we shall stop.
Likewise can we make the most safety critical bits less rigid. Often through linear movement held against an end stop by a suitable spring. Often eliminating most crushing hazards.
Safety is largely about estimating what will happen and when, and then stopping in a safe manner if we fall outside of our estimate.
And with feedback of this kind, one can often operate at noticeably higher speeds before crushing becomes an issue. (though dependent on reaction time and how long it takes the machine to stop moving.)
The 0.1 m/s is mostly an example of a safe operating speed for almost any equipment. It isn’t fast enough to crush bones or even leave a bruise. (unless concentrated down to a sufficiently small point.) Realistically, even 0.5 m/s is too slow to cause harm from getting hit. Crushing is however still an issue, but as I have stated earlier this can be solved.
-
Foldi-One says:
Not missing it at all, but no matter how slow you go the mass/stiffness/power of the robot arm can easily cause damage.
How fast something is moving is of no relevance to the harm it can cause you – going as near as makes no odds static relative velocity at impact could still kill you with ease with a crush against something else if its got enough inertia or driving force – though being that slow you probably die of hunger/thirst from the pinning action rather than being turned to jam by the rib crushing ‘unstoppable’ force. All going slower gives you the machine operator or the machines internal logic is more time to react, doesn’t take away from the force that can be applied, so the maximum impulse might be quite low, as the time to finish squashing all the softer fleshy stuff and really apply the full power is so long, which is safer, as you have time to react, and are likely just slowly pushed away – just of no use at all if you do get caught between it and something else and its not stopped quick enough.
Even if its got a superb position accuracy and shuts down as soon as it encounters forces just the tiny bit over the expected peaks of the sensors noise floor for this move in places it doesn’t expect to hit anything harm is possible, and while its actively expecting to pickup/putdown something the positional accuracy is largely out the window as the object usually isn’t 100% repeatably always held exactly in x position, as its a real world object – in this case the wildly differently sized and very non-uniform shaped chess pieces, so some resistance in the window where the part is expected to hit the board/grabber is 100% expected! To the machine that just means the part was rotated differently on the board or held very slightly differently in the actuator, not that something else got in the way…
(though get slow enough you’d have to insensitive to all stimuli to not be able to move out of the way as it moves to crush you)
-
Alexander Wikström says:
“How fast something is moving is of no relevance to the harm it can cause]…” is a largely unnecessary paragraph. Since yes a machine is obviously dangerous if it doesn’t stop…. While my whole argument is, “It shall stop if X condition happens.”
But going on.
Yes, inaccuracies in picking something up and placing it back down do exist. But saying, “there is inaccuracies!” makes you sound like a mathematician with no engineering experience. We should look at inaccuracies and consider if they are sufficient to be problematic, tolerance stacking and all.
For a lifting the typical chess piece, the run to run variance is likely quite small. (in my own experience I see little reason for it to venture beyond even 1 mm.)
As long as our variance is sufficiently small, then a finger is going to result in a considerably bigger variance.
In regards to positional accuracy. Being half the diameter of a small finger is sufficient. So 2-3 mm at most (to be safe to practically all ages). And most often we will have better positional accuracy than this.
Rigidity however needs to be better for better positional accuracy, and that in turn means larger crushing risks since the thing won’t give.
But yet again, the main crushing risk if when the arm goes down, a situation a spring loaded mount for the gripper would also solve crushing issues. Since even if it starts to clamp down onto the user, there is a couple of cm worth of fairly low force travel that the machine itself never considers using. (not that the machine in the incident had such.)
And with a spring loaded mount, it can have the end stop provide an electrical connection so that it notices if the mount were to move back against the spring. (ie, informing it that it shall stop.)
In regards to getting hit by a moving arm, this isn’t all that dangerous either. Low speed movements don’t impart a lot of energy. Even 0.3 meters a second is all things considered fairly slow, it will likely hurt. But damage is very unlikely. And 30 cm a second is quite rapid. (about a foot a second) Nor is the machine operating in an enclosed environment, so crushing up against something is likewise unlikely. (and it should still stop when confronted with unexpected resistance.)
However, the idea isn’t to make the machine completely incapable of inducing any form of harm. (because that is unrealistic) But rather making the machine incapable of inducing serious or even minor damage. It might in the worst case still inflict discomfort.
And in the end.
The whole concept I am talking about isn’t “new”, or unique.
A fair few industrial robots can do this, and do actually do this. If assembling two pieces and seeing forces outside of the norm, then a lot of industrial robots can alert the operator of the issue. Instead of potentially breaking expensive equipment that likewise can be hard to quickly replace. -
Foldi-One says:
And I don’t disagree with you – my point is NOTHING is without the potential for injury with these things, so some minor injuries from time to time are to be expected, no matter how safe its supposed to be and that this was a not a major injury… I’m not against robot arms etc or saying they can’t be perfectly adequate in safety, just that stuff can always go wrong and perhaps that is what happened here, despite sufficient safety precautions.
You seem to have rather more facts than I can find about the setup of this arm to speak so confidently it did or didn’t have x options…
But even if they had everything you suggested and nothing at all goes wrong my point is it STILL CAN cause harm. There is no such thing as foolproof safety – such as a kid putting their finger of comparable size to the expected target right into the spot the gripper is about actuate on, and that finger not being solid and slippery plastic but squashy flesh and tiny bone can’t take the clamping forces, or perhaps it can but then not being heavy and/or tough enough to take whatever movement happens before the robot automatically stops when the forces are unexpected.
Which for a prolonged period may well be the case as the forces are likely not going to be unexpected enough to trigger immediately, as in the real world sensors have noise and there is move to move and run to run variance as the grease in the joints changes temperature/ages and the parts wear in etc. Which gives you an operating window of expected that isn’t all that narrow if you want the machine to actually work rather than need resting every other movement.
Then remember as humans we are rather flexible and likely going to move with the robot in the hopes of avoiding injury and so exert very little force, till we suddenly do run out of flexibility to match a change of movement or can’t move fast enough around the chessboard in the way…
And as we all have already agreed fingers are delicate, moved in the wrong direction really really delicate. So even if you set the unexpected threshold pretty damn close to the point the machine never finishes a move without a reset quite likely enough to harm a finger. Even with everything you suggest to make the robot safe and no errors in the machine a finger injury is easily possible, it just takes bad luck, perhaps some ‘lazy’ programming assumptions (like nothing unexpected should ever be between the chess part and the actuator and no differentiating between tall and short pieces – so just telling the machine to go down till it makes contact, at which point a finger above a short game piece, a pawn say is within the expected height being still lower than the tallest part that is usually the Queen or King), then maybe add some stupidity/ignorance to create a human error too.
And all that is assuming nothing actually goes wrong in the machine…
-
-
-
Chris says:
Limiting speed may actually make it more dangerous.
No joke, one of my machines had to increase speed in its most dangerous point. It was slow enough that no one could hear or tell it was moving.
Recommend
About Joyk
Aggregate valuable and interesting links.
Joyk means Joy of geeK