Re: 1 Request and related point of discussion(with odd example)
- Hmmm, oddly enough the link worked when I previewed the post, but now it doesn't work for me either... The image is in the main files section on this board, just click on 'Files' and scroll down until you get to Nv Bias.jpg, or run a text search for it.
- Fantastic! Thanks!
It's just like what i have breadboarded with the exception of the diodes! This is great and simple to, kind like a wheastone bridge.
- Hey all,
Thanks again to telefox for the Nv potentiometer circuit.
And thanks again to Joseph Charles for the mibicore circuits.
I recently finished working on a new 5 motor walker with 1 new function.
For the Nv timing section I actually have 2 groups of resistors per Nv, with one group connected through ground by diodes and a switch, and the other groups connected normally.
The first section has timing resistors of value 2.2M and the switched section 1M. The switched section is activated by either a "bum" tactile sensor or an Ir detector placed on the rear of the robot.
When triggered, the timing is switched over to the 1M resistors and the robot will "run" away from what ever is behind it.
I was originally planning on using the spare inverter on the chip for an Nu neuron to keep the run function active, instead I used a capacitor in the switching circuit to keep the bot running for a while. After the run function is triggered, the robot gradually returns to its normal gait.
One could have multiple switched timing sections such as the 2 line system in the 5 motor walker, ti change the robot's functions, such as gait.
The next thing I am working on is a selection circuit ( hexcore) Which can select one set of timing lines for use, and cycle through the different timing lines until a particular one gives a desired result. Would it be possible to use Nu neurons (as in lobster) to "remember" the more successful timing line?
Would an Nv neuron referenced both to GND and VCC through mathced resistors suffice?
Thanks and Best Regards, Amit
- Hey Amit, the new bot sounds awesome. I'd love to see a demo video appear on your YouTube collection sometime.
Using Nu neurons to act as a short-term memory is definitely viable, and you can add a second positive feedback inverter if you want the memory to be more stable and long-lasting. The main problem I've encountered with trying to reinforce a certain neural network behaviour is how do you quantify what action is a success?
If you can figure out a circuit that allows the bot to detect how well it's performing in a given aspect, using memory to reinforce that behaviour should be fairly straightforward.
- Thank-you! I have to put in a new waist motor before I take the videos, but hopefully next week they will be up. For now, I can upload an old test video I made which shows the microcore tests of the bot.
Thanks for the tip regarding the second inverter, didn't think of it.
You are correct about the problem of quantifying success.
I think a solution lies in the robot's sensors, the robot needs to monitor itself in some way (my arduino glints at me...).
I have an idea to use a bicore, some Nu's and that neuron i had described earlier for that purpose.
Other than that, enforcing the memory should be easy, as you are just changing the probabilities of each timing line by stimulating particular Nu neurons more than others.
Thanks and best regards