Actions
Forum Thread Details
  • Replies 19 replies
  • Subscribers 405 subscribers
  • Views 1362 views
  • Users 0 members are here
Related

Thinking bots

Former Member

Hi! If your reading this then you should probably keep your mind open and just roll with what i am telling you. My team attends North East Magnet High School in Wichita KS. I have an interest in Control systems and electronics engineering. My teammate Austin Slater has always been adept at mechanical engineering and physical design. What we intend to do and in fact what we have already had some success with is designing, building, programming, and TEACHING. a humanoid robot. We are using neural network concepts to encode a Long-Short term memory neural network. This allows the robot to have thought and even memories in the form of "Physical" Connections between the "Neurons" and the signals being bounced around the "Brain". The bot has treads for mobility and is humanoid from the waist up. The bot has already learned to avoid sharp drops or "Cliffs" such as a table's edge. It has learned to make a punching motion and lean back to avoid being hit. (HINT. we built two bots to simulate a fight with each other). We still have farther to go but we are on a steady way. any help, advice, things you would like to see, or just cool comments, post them here. thank you

Parents

  • Ok. this is were you get into the murky field between robotics and psychology. The brain does store what we would coin "Useless" information. the same way you might remember going up stairs earlier this week. that memory has no purpose but it is there anyway. another thing that is unique about neural networks is "Compression" the deeper a neuron is from the "inputs" the more specific that neuron is. A study showed that one man had a neuron that fired every time that man saw the president. the neurons close to inputs represent simple things like lines, vowels, pressure on fingertips while the deeper neurons respond to more and more specific situations. the way my neural network decides what to store is the same as how your brain decides what to store. "Attention" and "Repetition" every time a neuron is used it gets a little stronger. So the more a action or stimuli is repeated, the more important that neuron becomes. eventually a long term memory of that stimuli is created via that neurons growth. the greater attention we give something, the longer it stays in our short term memory as signals that are bouncing around. that in turn also strengthens certain key neurons, increasing the chances of memory development. another key factor in neural networks is that memories are general. the same way many "simple" neurons work together to fire that "specific" neuron, a "specific" neurons triggers many "simple" neurons, which fill in the gaps of your memories. that "president" neurons triggers many simpler neurons, which fill in a face, a voice, a character. the exact interactions between neurons is a murky subject. the physical aspect of it is fully studied , but the way neurons interact over vast networks is a daunting subject at best. i hope i could shed a little light on the subject. let me know if you have anymore questions.

  • in reply to Former Member

    Ok I got how the brains stores info, but how do you do it.I mean Is your robot "editing" its code everytime it stores info.If yes where do you store that code, couse it would be big.And if i had to do it the first thing that i would do is make a ""base" code-one to be like an instict to that robot- which motor is his leg and stuff like that, and another part of that code to be the memory part, where the robot stores info and uses it later.Im I even close? image

  • in reply to dirtdiver

    Ok. you should read up on Neural networks. The neural network works by running many smaller class files called "Neurons". these classes mimic the function of actual neurons. one thing these classes hold is an array of "weights" or a value that is multiplied by the output of a neuron coming into this neuron. the array holds 1 weight for every neuron that outputs to this neuron. these weights are how the value is stored. you are correct in saying that all this data is large. I am currently running an Arduino uno with an Ethernet shield and using an SD card to store every single weight before shutting down. I am also using 4 I/O pins to connect to external Microcontroller RAM so I can store this data while running. I am looking for a better controller outside of the Arduino family due to it's lack of power. it is a great "MicroBoard" wanted to clear that up. the arduino is a microcontroller with a lot of add-ons. so I am looking at a 32 bit 144 pin 6.4 MHz microcontroller that will be more suited to my needs as this project is not a small one. my little arduino unit is sitting in my room at the moment. I won second place in a "Young Robotics" competition thanks to him. Now I will move into more powerful electronics that will allow Tron's (Tiny Robot On Neurology) big brother to run more smoothly. I will add tilt sensors, two colour camera's and more in order to give him greater versatility. not to mention how many control points he can have with 144 pins. thank you for your interest.

Reply
  • in reply to dirtdiver

    Ok. you should read up on Neural networks. The neural network works by running many smaller class files called "Neurons". these classes mimic the function of actual neurons. one thing these classes hold is an array of "weights" or a value that is multiplied by the output of a neuron coming into this neuron. the array holds 1 weight for every neuron that outputs to this neuron. these weights are how the value is stored. you are correct in saying that all this data is large. I am currently running an Arduino uno with an Ethernet shield and using an SD card to store every single weight before shutting down. I am also using 4 I/O pins to connect to external Microcontroller RAM so I can store this data while running. I am looking for a better controller outside of the Arduino family due to it's lack of power. it is a great "MicroBoard" wanted to clear that up. the arduino is a microcontroller with a lot of add-ons. so I am looking at a 32 bit 144 pin 6.4 MHz microcontroller that will be more suited to my needs as this project is not a small one. my little arduino unit is sitting in my room at the moment. I won second place in a "Young Robotics" competition thanks to him. Now I will move into more powerful electronics that will allow Tron's (Tiny Robot On Neurology) big brother to run more smoothly. I will add tilt sensors, two colour camera's and more in order to give him greater versatility. not to mention how many control points he can have with 144 pins. thank you for your interest.

Children