Shop OBEX P1 Docs P2 Docs Learn Events
Artificial intelligence in robots — Parallax Forums

Artificial intelligence in robots

hapaluahapalua Posts: 149
edited 2012-12-02 13:42 in General Discussion
Hello everybody. Does anyone here have extensive experience programming ai into their robots? If so tell me about it! :tongue:

Comments

  • BitsBits Posts: 414
    edited 2012-12-01 13:49
    I have experience but not as positive as I would like to admit. I have made a robot or two and always tried to code them to think on there own. I believe a robot without human involvement is the only way.
  • hapaluahapalua Posts: 149
    edited 2012-12-01 13:58
    Cool! What languages have you coded your robots in?
  • prof_brainoprof_braino Posts: 4,313
    edited 2012-12-01 16:16
    I don't have a lot of success stories either.

    But one method is to do a bunch of simple things consistently, and where they all happen together, it seems "intelligent". If you look at my current hero mark tilden's robots, they do a bunch of simple stuff, and come off as if they are actually "smart". And some don't even have a processor.

    Anyway, a good way to do a lot of simple stuff is in forth, its great for simple drivers and all that, and then string them together.

    For example, forth just does a bunch of really basic stuff, but folks often mistake it for an IDE or an operating system, since it gives you all the "necessary" bits that these provide. Its neither, but gives functions that allow one to do some of the things we associate with IDE or OS.
  • rjo__rjo__ Posts: 2,114
    edited 2012-12-01 16:34
    I am not entirely sure what artificial intelligence is anymore. It used to be an approach to parameterizing equations based upon feedback. The general problem is that the data in complex environments is never really the same... so any parameterization that your algorithm does is always dependent upon limited data and may end up not at all useful when the data changes a squish. What happens when you leave out a dependent variable... which isn't impactful during training, but becomes impactful during use?

    I personally like deterministic approaches... they may fail, but when they do it becomes obvious that there is a flaw in your thinking, and your thinking is then directed toward finding that particular flaw.
  • MacTuxLinMacTuxLin Posts: 821
    edited 2012-12-01 17:33
    Just a thought, another way of over-coming batch processing (train then implement on new dataset) is to use algorithm that handles streaming data. That way, your parameter (weights) will continue to adapt to new changes in your data thereby "skewing" towards the more favorable answer. Somewhat similar to what online web-site with continuous streams of data does.
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2012-12-01 17:53
    Google subsumption architecture.

    -Phil
  • mindrobotsmindrobots Posts: 6,506
    edited 2012-12-01 19:37
    Subsumption architecture & sensor fusion

    http://www.solarbotics.net/library/pdflib/pdf/BattlingReality.pdf
  • ElectricAyeElectricAye Posts: 4,561
    edited 2012-12-01 22:00
    Phil Pilgrim (PhiPi) wrote: »
    Google subsumption architecture.

    -Phil

    AKA Forrest Gumpsumption architecture. :)
  • hapaluahapalua Posts: 149
    edited 2012-12-02 10:36
    Who here agrees that JAVA is the best programming language
  • Heater.Heater. Posts: 21,230
    edited 2012-12-02 10:45
    hapalua,
    Who here agrees that JAVA is the best programming language

    That's dangerous talk in these parts.
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2012-12-02 10:53
    hapalua wrote:
    Who here agrees that JAVA is the best programming language
    heater wrote:
    That's dangerous talk in these parts.
    ... and the wrong thread in which to begin such a discussion.

    -Phil
  • mindrobotsmindrobots Posts: 6,506
    edited 2012-12-02 11:22
    Back to the OP's question.....

    AI is a pretty broad subject. What types of things were you wanting to do that you consider AI??
  • NWCCTVNWCCTV Posts: 3,629
    edited 2012-12-02 11:54
    There is a thread around with video and all that discusses this The OP had his bot "learn" from it's mistakes. Kind of cool. I will see if I can find the thread and link to it.
  • Duane DegnDuane Degn Posts: 10,588
    edited 2012-12-02 13:42
    NWCCTV wrote: »
    There is a thread around with video and all that discusses this The OP had his bot "learn" from it's mistakes. Kind of cool. I will see if I can find the thread and link to it.

    I looked for that thread myself (I couldn't find it).

    I think the short coming of that approach is it required a second set of sensors to detect "mistakes". I still thought it was very cool. I'm pretty sure the guy used some sort of Basic Stamp as the controller.

    I believe he used terms like "pain" to describe sensor input that lead to mistakes.

    I hope you find it.
Sign In or Register to comment.