As blogger wont let me upload more pictures I will take some time to think about what I have done so far...
What are the main features of my prototype?
- It is modular. There are basic Velcro pads which form the structure for mounting the sensors. The sensors are then placed on the Velcro.
- The sensors are off-axis. Because of this, only one side can be permanently mounted to the body. The other half is secured by small pipes through which the cables can freely move back and fourth.
- The Arduino will be mounted to the chest and will communicate the data to a computer. (For now via USB. It might by nice to have it directly sending data to the internet using WiFi or Xbee)
How does that work out?
I’m not quite sure. The off-axis location of the sensors and the modularity make for a simple and “soft” construction. This is probably much more comfortable than a rigid construction. Getting the potis exactly on the axis of rotation would be quite hard, especially if the interface is intended to be worn be different people. Having said that, it would greatly increase accuracy and create a more stable construction. This would also make feedback processes easier to incorporate.
Why did I decide for this particular design, instead of say, optical motion capturing?
- Having a one to one correspondents between sensors and actuators minimizes the processing required before the signal can be used. This should make the software which will be needed simpler and possibly faster. Ideally the delay between performing a movement and the Nao copying it should be neglectable.
- A physical capturing system can potentially give the most precise info on what is actually happening, irrelevant of the surrounding conditions. (Try using an inertial system on a train. Try entering an elevator with an electromagnetic device. Leave the proximity of your external sensors for ultrasonic and visual systems.)
- A physical capturing system can potentially be upgraded to incorporate feedback.
So, does this prototype actually justify the method I chose?
I would say yes. Most of the objections I have found in literature and heard from my professor and other people I have discussed this with have been solved. The modular setup allows it to be easily adjusted to be usable by anyone, irrelevant of the person’s size. Calibration will be necessary before use, but is quite simple. Later on I hope to actually automate the calibration process. The pads are neither heavy, nor uncomfortable: I was wearing them for several hours yesterday, while experimenting with the calibration. Except for the fact that I was attached to the computer and thus not exactly mobile it neither hindered me working, nor was it especially uncomfortable.
Most important of all: It works.
Does my prototype fulfill the three requirements I originally had? Well… only to a limited extent. The one to one correspondence is given. The precision however can still be greatly improved; especially the rotation of the upper arm can be improved upon. Possibly stretch and flex sensors will be able to do the trick. Upgrading it to incorporate feedback is not possible as of now. I would like to add force feedback by attaching servos to the joints, this however would call for a sturdier construction.
The problem I see here is that, if I decide to use non-rotational sensors, incorporating force feedback becomes extremely tricky. It definitively would lose the beautiful simplicity I see in the one to one correspondence. (I still could go for tactile feedback though… … anyway feedback is a far away idea so I will stop worrying about that now.)