Building humanoid similar to Asimo and Qrio
by siempre.aprendiendo
Tue Sep 18, 2012 11:57 am
Thinking in the function, the sensors and the software needed... I would vote for the 1) 2DOF fingers.
I think that it would be easier, and probably more successful, to grasp an object with 2DOF than calculating the correct hand orientation and trying to grasp the object with only 1DOF for all the fingers.
But developing software for using Asus Xtion to perceive and to control the wrist could be really interesting, and hard
I think that it would be easier, and probably more successful, to grasp an object with 2DOF than calculating the correct hand orientation and trying to grasp the object with only 1DOF for all the fingers.
But developing software for using Asus Xtion to perceive and to control the wrist could be really interesting, and hard
by limor
Wed Sep 19, 2012 10:54 pm
we spent the day building the leg and discovered lots of issues related to building. Some parts take a lot of screws in difficult places, some need to have designated bolt slots added so that they don't interfere with the surface underneath.. all good because the fast 3D printing cycles of design-print-assemble help to perfect the design.
by limor
Fri Oct 05, 2012 4:16 pm
I've been traveling quite a bit the past weeks so the robot hasn't evolved much. However, I've been showing it around to people at conferences I attended and got great feedback. People love the mechanical look and the fact that this is almost completely 3D printed.
Eventually this will be the first humanoid robot that is fully 3D printable (except for electronics, servos, screws, carbon fiber rods, kevlar wire and maybe some aluminum around the central body).
I've been toying with the idea of eventually selling all the non-printable parts to people with access to a 3D printer and they would then print and assemble the robot. Printing the parts will probably take 2-3 days.
The other interesting advantage of this approach is that whenever we do a structural improvement of the robot, we can share the new models with those who already own the robot and they can "print the upgrade".
Any feedback welcome.
Couple of improvement already here:
Turns out the thin kevlar wire is strong and can do the job well but it is unexpectedly difficult to clamp onto the 3D printed pulley. It is just too thin and slippery. We even tried to screw it into the plastic, and use epoxy but results are not consistent.
The solution is to use a thicker kevlar wire. Same sports equipment shop (decathlon) sells also fishing wire intended for catching sharks and whales. ie: strong enough for this robot and obviously thicker. The advantage is that it has a wider surface area and can more easily be clamped down onto the pulley without slipping over time.
Another thing that was unexpected was that wig bracket and knee bracket require a lot of screws, screwing and fidgeting. Also, there is space between the servos that increases as force is applied in various directions and risks deforming over time.
.jpg)
To address this, we created a double wig bracket holding 2 servos at a time. Here's a first attempt, simply using sketchup to merge 4 brackets into a single entity. Proper CAD design will follow.
.jpg)
Eventually this will be the first humanoid robot that is fully 3D printable (except for electronics, servos, screws, carbon fiber rods, kevlar wire and maybe some aluminum around the central body).
I've been toying with the idea of eventually selling all the non-printable parts to people with access to a 3D printer and they would then print and assemble the robot. Printing the parts will probably take 2-3 days.
The other interesting advantage of this approach is that whenever we do a structural improvement of the robot, we can share the new models with those who already own the robot and they can "print the upgrade".
Any feedback welcome.
Couple of improvement already here:
Turns out the thin kevlar wire is strong and can do the job well but it is unexpectedly difficult to clamp onto the 3D printed pulley. It is just too thin and slippery. We even tried to screw it into the plastic, and use epoxy but results are not consistent.
The solution is to use a thicker kevlar wire. Same sports equipment shop (decathlon) sells also fishing wire intended for catching sharks and whales. ie: strong enough for this robot and obviously thicker. The advantage is that it has a wider surface area and can more easily be clamped down onto the pulley without slipping over time.
Another thing that was unexpected was that wig bracket and knee bracket require a lot of screws, screwing and fidgeting. Also, there is space between the servos that increases as force is applied in various directions and risks deforming over time.
To address this, we created a double wig bracket holding 2 servos at a time. Here's a first attempt, simply using sketchup to merge 4 brackets into a single entity. Proper CAD design will follow.
by limor
Sun Dec 23, 2012 1:58 pm
It's been a hectic end of year for RoboSavvy and I've been too busy with expanding the business. Finally some time off now and I intend to finish this project.
The two most challenging things that we faced with this architecture were (a) the center structure was done in sheet metal and (b) the wire based transmission has clamping issues given the limited resolution and rigidity of 250 micron 3D printing in ABS.
So the first challenge I'm working on is to make the whole structure 3D printable. The aluminum sheet cutting that was previously there is challenging, expensive and it requires complex bending. Here's my first take on 3D printable torso.
creating a 3D printable bearing will require some experimentation because the clearance between the two touching sides of the rings must be minimal
if it doesn't work, then i'll have to use metal bearings or injection molded rings of nylon or delrin.
The top shelf is quite complex because I want a 1 to 2 gear reduction between the AX12 and the top shelf (not shown here).
The upper shell will be laying on top of the bottom shelf and there will be need for support on the wide sides to remove any wobble
The battery may go in the groin space between the legs. Specifically I'm thinking of this battery spec'ed 1800 MAh with about twice the juice of the Bioloid.
Here's a picture from behind. That's a Rasberry-py on the back which will need some shielding and positioning.
The two most challenging things that we faced with this architecture were (a) the center structure was done in sheet metal and (b) the wire based transmission has clamping issues given the limited resolution and rigidity of 250 micron 3D printing in ABS.
So the first challenge I'm working on is to make the whole structure 3D printable. The aluminum sheet cutting that was previously there is challenging, expensive and it requires complex bending. Here's my first take on 3D printable torso.
creating a 3D printable bearing will require some experimentation because the clearance between the two touching sides of the rings must be minimal
if it doesn't work, then i'll have to use metal bearings or injection molded rings of nylon or delrin.
The top shelf is quite complex because I want a 1 to 2 gear reduction between the AX12 and the top shelf (not shown here).
The upper shell will be laying on top of the bottom shelf and there will be need for support on the wide sides to remove any wobble
The battery may go in the groin space between the legs. Specifically I'm thinking of this battery spec'ed 1800 MAh with about twice the juice of the Bioloid.
Here's a picture from behind. That's a Rasberry-py on the back which will need some shielding and positioning.
by limor
Wed Dec 26, 2012 7:48 pm
playing around with the position the of 200g battery vs. the RPi module.
additional electronics include a UBEC voltage regulator to feed the RPi, a tiny wifi module on the RPi, IMU, and USB2Dynamixel or arduino pro mini as interpreter between the Dynamixel bus and the RPi USB. Not sure if the Robotis software suite for open-loop control should even be used here. this is a brand new thing which should radically break the open-loop paradigm.
additional electronics include a UBEC voltage regulator to feed the RPi, a tiny wifi module on the RPi, IMU, and USB2Dynamixel or arduino pro mini as interpreter between the Dynamixel bus and the RPi USB. Not sure if the Robotis software suite for open-loop control should even be used here. this is a brand new thing which should radically break the open-loop paradigm.
by limor
Mon Dec 31, 2012 12:49 am
I've added legs and head.
The head has 2DOF deploying similar configuraion as for arms and legs
Total of 28 servos. I'm planning to create further reduction in some of the joints where wires are used, by putting small/big pulley. Lets see if this baby can walk
(the head is a placeholder. I don't really plan to put a Primesense sensor in the head. but dual web cam probably yes)
The head has 2DOF deploying similar configuraion as for arms and legs
Total of 28 servos. I'm planning to create further reduction in some of the joints where wires are used, by putting small/big pulley. Lets see if this baby can walk
(the head is a placeholder. I don't really plan to put a Primesense sensor in the head. but dual web cam probably yes)
by limor
Mon Dec 31, 2012 6:32 pm
first attempt to print the torso's complex part was nothing to brag about.
once support material is involved in a precision part, things don't usually have a happy ending. i'll cut up the part into two individually printable parts and then will have to screw the resulting parts into a sinble piece. the circle has little facets rather then being smooth. maybe the STL needs to be saved with higher resolution.
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
once support material is involved in a precision part, things don't usually have a happy ending. i'll cut up the part into two individually printable parts and then will have to screw the resulting parts into a sinble piece. the circle has little facets rather then being smooth. maybe the STL needs to be saved with higher resolution.