Bipeds

[nxt1976]

he doc,
Nice work, it's worldclass! I'v been watching your biped project for a while and I can say that your progress is awesome.
Keep working on that dream, I'm sure you'll make it true!!
greetings LC

[dobbers]

Hi Doc,
Something puzzled me after looking through this topic. Do you have sensors on the joints to feedback what angle they are position at? I was wondering how you would "reset" the legs to a known position to start the walking process again after a stop mid flow.

Great work BTW.

regards,

dobbers

[doc222]

Each Lego Motor(servo) has a rotation sensor built in. The NXTmx's used also read this as well. So setting it back to starting possition or any possition is possible. Though in testing I have found at times not always exact. I had to replace a motor the other day seems that the rotation sensor may have died, thoughI have yet to test this in another way other than just in X. I tried a new cable it still would not read. I replaced it with a extra motor I have and it seems ok now. So may very well have died or perhaps the motors plug where the nxt cables goes may be worn. They are getting old, all my motors are over 3 years old now.

Right now I am working out what is the most efficient way to walk, there are many ways to walk for a robot it seems. When this is worked out then I hope to be able to scale steps from short to long steps depending on where it is and readings from other sensors. So far the way we wlk is best for the least amount of energy as compared to the "Chicken" walk. But when climbing high steps the chicken is best. Mainly a mechanical advantage with the design i think. Since it is not a humanoid really just apair of legs with a head at the belly button height . To step up say a 5 inch step means ALOT of leaning to get over the feet and if the knee is forward its in the way, put the knee backwards and its not. But the build seems very solid right now, even these Lego La's seem to be working very well and have no issues with the amost 9lbs of bot. Very heavy.

[doc222]

X-2.6a my on-going project, latest version. Looks like it will be able to climb the almost 5 inches steps and be able to climb down as well. Though the program is not polished enough to do it with out gravity finding a weakness , all the moves are there, just needs afew more hours of tweeks. One thing about lean as compared to shifting the weight by sliding it back and forth. IT is WAY harder to lean side and side. 5inch steps is the most this biped will be able to climb, even to do this to tilt over the stepping foot the other foot has to tippie-toe to get it too roll over the leading foot. Once again like doing shorter steps the "going down steps" is another world of hard to get right at 5 inches it is almost scary(if you seen x-2 having a bad day bloopers you know how this can end). All this has made getting this done very long process. Early versions of this bot just going up these steps lead to some rather disastrous falls. It had the rack weight shifting that made it a bit easier then now it must lean to stay over the leading foot. No film yet, but looks like very soon.

Here is pic of it doing the long step down.

[h-g-t]

That is an amazing bit of kit! Well done.

[doc222]

X-2.6a has completed the task of "5 inch steps" up and down. X-2 was never designed to climb things of this height.
X-2 is only about 20 inches tall and weighs about 9LBS. Climbing up or down these size steps is pushing this desighn to
its very edge. It has once again took a lot longer to do this than I thought, proving again that nothing is simple when
it comes to robotics and Bipeds. I have been thinking on what it will take to get a form of AI feedback, Ie sensor to control
and monitor walking, climbing etc. For one it would need a few sensor mux's. It also seems that about all these sensors would
need to be in the feet. Since we get all our information feed back from ours. Imagin how hard it would be to walk if you could
not feel the floor? Walking would become visual only and and best ataxic. So now I need research sensor muxs, i figure a touch
mux and a degital should cover it? Though in the end with out having any in hand its all guessing.

Each foot I think would need 2 touch, one gyro and eceleration to start I am guessing. Up near the head a very accurate distance
sensor I am thinking too, one of the medium range IR from mindsensors. This to scan and give details of objects in its path or to
measure steps height or decents. Like any project though it takes funding and a budget.

One thing X2 lacks in design is a way to rotate its hips. Looking at Asimo, it has this ability in each leg at the hips; to twist a bit or rotate.
I think in Lego this will be rather hard to implicate as done in Asimo and keep it compact and not to loose. I have an Idea
that will give this freedom of movement but uses one one DOF as compared to Asimo's two Dof that lets it turn with out
"having to shuffle feet." The basic idea is to hinge the hips at the point between the two legs. This will let the leg that is not touching the ground
to twist forward or back. I have yet to even build this new hip cage, it is all in my mind at this point and well what one thinks may work in life may not.
Only testing and protyping will prove if the idea has merit or is just another "mind vapor ware".


X-2 on the 5 inch stairs

[h-g-t]

I am even more impressed now that I have seen it in action!

[doc222]

X-2b 11 dof

X-2a's1: After working out different walking gates and turning patterns. I think that I found an optimum walking pattern for least amount of energy for the given amount of cycles it can make. One would think that the longest step would be faster but since at a point you must counter the action or go into a dynamic gate. Since going dynamic seems out of the question with the platform and staticly once you step to far you must counter this wth an opposite action. The gate then is one that can be done with the least amount of counter actions. This makes the cycles faster since in the end its just putting one foot in front of the other. Something like this has seemed to be best



This brought up a new issue now that I do not like. The way X-2a's version can not rotate/turn mechnicaly. To fix this issue I have made it turn using the "shuffle" as seen in the video. Works but only on smooth floors and if a foot catches or friction varies then well...not so good.

Programming/ software fixes for a mechanical deficiency is a poor solution so I have thought how one could fix the issue. Back a few years ago, x2(little x smaller bot had feet that rotated. Worked very nice but that was a smaller lighter bot, that only had to walk. Other lego bots have used this in the past. Asimo can rotate its hips enough to have a turn as well. Both these would need two more dof. What I wanted was 1 dof and to be able to turn left or right.

What I have came up with is simple in thought and build, though a bit harder to replicate in Lego and keep it strong enough to hold the forces at hand. Basicaly the idea is to have two hips, a left and right. These hip pods are then hinged, much like a door hinge. A Linear ram opens and closes the hip podes. So fare in mechanical testing this works well and it can now do a 45* turn in one step. so two steps should be very close to a 90* turn. It took 5 cycles of the "shuffle" to get to 90* before. Pluss you can not shuffle on carpet or other frictionous surfaces.
Hip closed

Hip open

side view
Testing it in worst case senarios, it is best to only use this hip open command in a vertcaly standing possition. If say one hip is at 90 degress to the other hip. Like standing on one foot and leaning forward until you are what looks like a "T". opening the hips will cause the Linear to over load way over its abilty and the whole hip assembly could fail. but Locked closed in these weird possition is ok.

regards

Doc"moving staticaly forward one brick at a time"

[doc222]

X-2b

I have been testing this a bit with no extra power supply. The weight of 12 AA'a is alot, though they are also balast to counter the weight of the legs. I am VERY suprised and rather pleased too.
As part of X2's upgrades:
(11 dof now, turning hip) This has worked of very well

Moded NXT with power out: I have run 8 motor commands at once all motors seemed to run very good. I did see all most 1volt drop during cycling the 8 motors. Have not ran all 11. In reality though there is very few times this would happen if every and many of the dof's may be just running 1-3 revolutions at atime as well. So it seems that the NXT can run other things like 4NXTmmx. I wonder how many motors at peak draw it takes to brown it out even with fresh batteries? X2 and every biped i have worked with has one week link its the knees. The knee puts the most load of all areas by far. So X2 has only two dof that at times may draw highest peak(ex. deep knee bends on one leg climbing 5 inch steps etc.) So this mode may work for me, time will tell.

Working on the foot plate X2b: "smart foot"
Old X2's foot plate before was just old 16L studded beams pegged and plated solid. since its a square thats about 16 beams. I removed every other and added some studless parts, skeletonized. The foot plate has hard points now for adding sensors and is lighter. I wish I could have each foots corner have a pressure sensor. I think for a biped this would be needed input for balance as much as a gyro or exl. But what I am trying is to get one sensor do two funtions it could be done with a touch or light sensor. To feel the front edge of the foot if it is on the ground and feel the front if the foot touches anything and not add to much weight to the feet. I like the light sensor Idea best since I could use color to tell if its ground or obstacle. Design is just a front crash bar when pushed it blocks the light sensor, "ground" is a pin/axle that drops down if the foot is lifted this removes a colored paper target. If both are present though it will give a grounding signal, this is an issue "mechanical" that could be fixed in programming i think.

[aswin0]

Each foot I think would need 2 touch, one gyro and eceleration to start I am guessing. Up near the head a very accurate distance
sensor I am thinking too, one of the medium range IR from mindsensors. This to scan and give details of objects in its path or to
measure steps height or decents. Like any project though it takes funding and a budget.

Doc,

What do you exactly want to use the sensors for? I presume it is to find out if the foot is on the floor or not. This can be done with force sensitive resistors or with flex sensors.

Also I would not use sensor multiplexers. If you use digital sensors you can chain up to 127 sensors to one sensor port. Just make sure that you can change the I2C address when you buy a sensor. (Hitechnic sensors do not qualify).

I still love your robot.

Aswin