Hardware build record

daylanKifky

Yeah, nylon is better. Non magnetic metal should work, but we had reports of users having problems with that. Since you have already mounted these ones it would be great if you can perform a calibration now and another one after you change them and compare the results.
If you do that please paste here the calibration offset and matrix that the notochord prints after the procedure

cubesky

daylanKifky
Ok, I will do this when I get them from @YJBeetle . It maybe 2-3 days later, because we don't in the same city, K-Ceptor and case need express delivery.

cubesky

It there any way to connect custom i2c device to hub (with address transforming like K-Ceptor) and get data from it during notochord running?
@daylanKifky

daylanKifky

Absolutely, and I'm glad you asked because the system is designed to accept other type of sensors, but until now nobody has actually used it this way

A good idea would be to use the same RJ12 connectors the KCeptor has, so the new device can be daisy chained with the rest of nodes.

Now, at the moment in the software side things are a little tricky, you would need to directly modify the notochord's C++ code. Give me more details about what you are trying to achieve and I will guide you.

We are also currently developing the new version of the notochord in the form of a python module which wraps most of the internal units of the notochord and will give you a much greater flexibility and ease of use for this kind of tasks, but if you want to try it now modifying the current, pure-C++ version is the way to go

cubesky

daylanKifky I want to use flex sensor to track finger bending, and send them to render software. Using the same RJ12 connector is great, I just search on notochord source code to find how to add custom device, but nothing found.

daylanKifky

cubesky I just search on notochord source code to find how to add custom device, but nothing found.

The thing is that the current notochord is an old dinosaur. It was the first program to be created for the Chordata framework several years ago when the actual scope of the project was not really defined. It then grew during the years to incorporate the different required features without a predefined plan. The result is a somewhat messy codebase that is hard to expand and maintain even for us. That's why we are practically rebuilding it from scratch with the new notochord module.

That's to say: even if the architecture it's prepared to incorporate custom sensors, it might be tricky to add the required code to the notochord

You should start by creating a class derived from Chordata::I2C_Node as the Chordata::K_Ceptor class in Chordata_node.h.
This class should override the virtual void bang() method.

Upon startup the notochord creates an instance of the Chordata::Timer and configures it with the required sensor read rate. It then fills it up with the different types of nodes in the XML hierarchy, for now the nodes are of type K_Ceptor and Branch. They both expose the bang() method which includes the code required to make the read/write and data processing of the nodes . The Timer takes care of doing this action at the correct rate and order to open/close the correct gates and walk all the hierarchy tree.

In order to get your custom nodes into the Timer you would need to modify the Chordata::Armature_Parser::parseNode method creating the code specific to your case.

Last but not least, the XML is linted against a schema, any unexpected XML structure passed to the notochord will result in an error and a halt in the execution. In order to add a new type of content in the XML you would need to declare it in the Chordata.xsd file which is of the XML schema definition for the input XML. This is a quite specific language written in XML, perhaps the easiest way to go if you are not familiar with it is that if you get to this point just tell us how you want to structure the new XML and we apply the required changes to the XSD

cubesky

I found an amazon link for flex sensor.

cubesky

I think we need calibration process to calculate resistance value when hand fully closed and fully open and then write it to eeprom.
We decided to use a 16-bit chip with ADC to get flex sensor resistance value and calculate these value to a 0-255 value and send it to notochord. Or we can send raw value and eeprom calculate data to notochord, just like K-Ceptor.

daylanKifky

daylanKifky
Another option would be to start applying this changes directly to the new notochord module. Everithing should be much more straight forward, but the module is still under heavy development. It lacks a few features to be able to perform a proper full-body capture as the implementation of the Timer. We are now working on it expecting to release it in the following months

cubesky

daylanKifky
Great, I will take a look.
But I don't understand how to calculate and set I2C address to notochord as we need. It seems I2C Address is related to LSM9DS1 and I2C Address transform chip now.

daylanKifky

Yes, the LTC4316 performs an XOR operation on the i2c address,

So for example if you want to read the mag which has addr=0x1E and you set the translator to value 0x02 from the outside you will see the mag as if it had the address 0x1C

cubesky

daylanKifky
So, if we build a custom sensor, we also need our i2c address on 0x6A/0x1C with LTC4316 to change address to (address XOR id) to avoid address collision with exists K-Ceptor. Is that right?

cubesky

I just receive express from @YJBeetle and test screws with following types.

Iron Screw

<calibration type="offset" target="mag">
                -1357 -1427 -2338
</calibration>
<calibration type="matrix" target="mag">
                0.982179 -0.0334575 0.0119249
                -0.0334575 0.976424 0.0206434
                0.0119249 0.0206434 1.04455
</calibration>

Stainless steel Screw

<calibration type="offset" target="mag">
                -114 -856 639
</calibration>
<calibration type="matrix" target="mag">
                0.977082 -0.0388209 0.00715508
                -0.0388209 0.988014 0.00368313
                0.00715508 0.00368313 1.03756
</calibration>

Nylon Screw

<calibration type="offset" target="mag">
                -238 -803 653
</calibration>
<calibration type="matrix" target="mag">
                0.974399 -0.0390949 0.0194447
                -0.0390949 0.988392 0.00872031
                0.0194447 0.00872031 1.04046
</calibration>

cubesky

Assemble the case with K-Ceptor.

daylanKifky

Thanks!
It seems indeed that the stainless steel and the nylon ones are giving similar results, so in theory they both should be safe to use.
In case of doubt use plastic as long as you can

cubesky

Temporary sensor mounting solution. Sorry, my camera is not good.

Today, I sewing "velcro" for 7 K-Ceptor using about 4hrs. Another 8 K-Ceptor include hand and head node need time to wait more "velcro" to arrive(It is used more than I expected.), I think I will transform a hat and a couple of gloves to mount K-Ceptor on hand and head.

The first proof of concept pcb for finger sensor is submit to pcb prototype manufacturer. I will release the design of the pcb for open source(It's super simple now), if it works with notochord.

cubesky

I just test for r-thigh r-leg r-foot l-thigh l-leg l-foot base. Other node is comment out in xml.
I kick forward my leg, but it seems wrong.

daylanKifky

cubesky I kick forward my leg, but it seems wrong.

did you do the pose calibration facing south?

We have recently developed a new pose calibration algorithm that should behave much better with this type of problems. We talked about it in the last livestream

cubesky

daylanKifky did you do the pose calibration facing south?

No. I will try again tomorrow with full default biped configuration.

cubesky

I fixed all 15 K-Ceptor to my body and calibrate it facing south.

It seems not correct, leg and arm is move to the wrong side.

By the way, I don't know where should I fixed the K-Ceptor. It should fixed to the side of leg/arm or the front of leg/arm. I can not find any instruction on wiki(also not found in video above). But I test front and side to my leg, but it also move to wrong side.

During this test, I found two K-Ceptor jump to wrong ID number due to resistance soldering issue. I make a discussion with @YJBeetle ,we think maybe use CAN bus with pre-defined unique id (Such as chip id) by device instead of I2C, and config it with xml is more better than use resistance to change id. Select ID by using voltage is not robotic. I2C is more usable when chip communication on the same board, long wire may cause I2C communication error by noise. CAN bus is more robust, Linux also support it and CAN bus provide star-style link function, all K-Ceptor can run on the same bus without use mux chip (like PCA9548AD) and It can support a very large number of nodes on the same bus.

Due to IC chip shortage, finger sensor is delay, I use ADC128D818 to convert voltage measure data to I2C, but its price raise from 14 RMB (about 2.2 USD) to 320 RMB (about 50.25 USD). We need change this chip to another chip, such as ADS7828.

daylanKifky

cubesky By the way, I don't know where should I fixed the K-Ceptor. It should fixed to the side of leg/arm or the front of leg/arm. I can not find any instruction on wiki(also not found in video above). But I test front and side to my leg, but it also move to wrong side.

It doesn't really matter where in the body segment you fix the sensor. You just need to find spots where the soft tissue (muscles, fat, etc) is minimum and the influence of the rigid bone underneath over the surface of the body is greater. IN some parts of the body such as the upperarm it's hard to find a good spot. In others like the head the influence of muscles over the surface is minimal.

« Previous Page Next Page »