Artificial Intelligence - Robot

Thread Starter

Pushkar1

Joined Apr 5, 2021
401
I have been reading article on Artificial Intelligence for the past three days but I really don't understand how it works. I understand that most embedded systems use structure based programming whereas I think AI is different. I think AI use knowledge based programming technique. AI system needs training or knowledge for better results. We must have input data to create model for the AI system if we do not have input data we cannot process further. The more the input data set, the more accuracy we will get. It's all theoretical stuff that I've got.

I want to understand how AI works through an example so I choose one example which is Room Cleaning Robot System. I have chosen the topic because it is related to electronic and there is nothing special behind it.

So let's assume that we have a room cleaning robot which is used for cleaning the room. I know this system can be programmed through structure programming. In this, the robot checks each condition and executes its tasks according to conditions.

My question, Is there any benefit to using AI technology rather than a structure programming for this robot project?

The purpose of this thread is to find out how AI can be useful instead of structure programming. @nsaspook I have read many of your posts related to AI. hope you can give some thought's on this
 

xox

Joined Sep 8, 2017
608
I have been reading article on Artificial Intelligence for the past three days but I really don't understand how it works. I understand that most embedded systems use structure based programming whereas I think AI is different. I think AI use knowledge based programming technique. AI system needs training or knowledge for better results. We must have input data to create model for the AI system if we do not have input data we cannot process further. The more the input data set, the more accuracy we will get. It's all theoretical stuff that I've got.

I want to understand how AI works through an example so I choose one example which is Room Cleaning Robot System. I have chosen the topic because it is related to electronic and there is nothing special behind it.

So let's assume that we have a room cleaning robot which is used for cleaning the room. I know this system can be programmed through structure programming. In this, the robot checks each condition and executes its tasks according to conditions.

My question, Is there any benefit to using AI technology rather than a structure programming for this robot project?

The purpose of this thread is to find out how AI can be useful instead of structure programming. @nsaspook I have read many of your posts related to AI. hope you can give some thought's on this
Well of course AI is much more powerful than the "standard" approach. The tradeoff however is complexity. Unless you're talking about a (most likely expensive) pre-packaged solution that provides some sort of API to interact with then you'll be doing most of the grunt work yourself. Which could well mean quite a bit of study on your part in order to become comfortable with all of the theorems and equations involved. So not exactly for the faint of heart!

AI isn't one single technology either. There are lots of different methodologies in use and some have even been around for a VERY a long time. Logistic regression and Bayesian statistics for example. So really if you get right down to it it's just a matter of familiarizing yourself with those techniques then figure out which of those can be applied to your specific problem-domain.
 

Thread Starter

Pushkar1

Joined Apr 5, 2021
401
AI isn't one single technology either. There are lots of different methodologies in use and some have even been around for a VERY a long time. Logistic regression and Bayesian statistics for example. So really if you get right down to it it's just a matter of familiarizing yourself with those techniques then figure out which of those can be applied to your specific problem-domain.
what I have found so far, AI is technique and Logistic regression and Bayesian are mathematical algorithms. For room cleaning robot we have to make a model in which we need a lot of input data to train model.

Lets assume we have a room cleaning robot that has a camera and it use the camera to see garbage and dust. Wherever the robot sees this dust, garbage it cleans the place.

I want to understand difference If I use AI technology instead of using standard method how will it be beneficial for robot. how the AI program will work for the robot, how the robot will be trained.
 

nsaspook

Joined Aug 27, 2009
8,898
No future intelligent machine will use only knowledge based or traditional programming. A blended approach will likely be used. Narrow AI is powerful as a classifier of objects but the physics to handle interactions with the objects identified are likely to be optimized using traditional computing techniques. If you need to detect dog turds and have a large library of dog turd images the AI based computer vision detection of turds should be beneficial for the robot to reduce the chances of scat polished floor.

Today's AI system can also be easily tricked or manipulated to be 'counterproductive'.
https://www.straitstimes.com/asia/e...yq-and-xiaobing-re-educated-after-rogue-rants
BEIJING/SHANGHAI (REUTERS) - China has taken down two online robots that appeared to go rogue, responding to users' questions with one saying its dream was to travel to the United States and the other admitting it was not a huge fan of the Chinese Communist Party.

The two "chatbots", BabyQ and XiaoBing, are designed to use machine learning artificial intelligence (AI) to carry out online chats with humans. Both had been installed onto Tencent Holdings' popular messaging service QQ.
 

Thread Starter

Pushkar1

Joined Apr 5, 2021
401
No future intelligent machine will use only knowledge based or traditional programming. A blended approach will likely be used. Narrow AI is powerful as a classifier of objects but the physics to handle interactions with the objects identified are likely to be optimized using traditional computing techniques. If you need to detect dog turds and have a large library of dog turd images the AI based computer vision detection of turds should be beneficial for the robot to reduce the chances of scat polished floor.

Today's AI system can also be easily tricked or manipulated to be 'counterproductive'.
https://www.straitstimes.com/asia/e...yq-and-xiaobing-re-educated-after-rogue-rants
I want to understand difference If I use AI technology instead of using standard method. how will it be beneficial for room cleaning robot.?

how the robot will be trained in my example?
 

Thread Starter

Pushkar1

Joined Apr 5, 2021
401
It might help if you pick an individual feature of the AI vac and compare it to how it is traditionally done.
So I have done more research about it and I think that AI should be used when there are many large number of inputs conditions.

Assume I have made a house cleaning robot for my home by the traditional method. I programmed it in such a way that it can only clean my house. if i want to make universal robot like it clean your house, hospital, school airport i think i should use AI. I should train the robot in such a way that it can clean any place through AI

i dont know if i am going in the right direction i have explained everything i just posted what i understood after reading on the internet article. I hope someone corrects me if I'm wrong
 

nsaspook

Joined Aug 27, 2009
8,898
AI is mainly a hype word.
Actually this is a good article but my basic question is why developers are moving towards AI from standard approach? Where does the standard approach fail or what is the benefit of the AI? For example what are the benefits of using AI for a room cleaning robot?
There is no real advantage other than hype for selling a product. AI today is just a different way to program in the standard way. It offers no real intelligence advantages because there is no intelligence.
https://billdembski.com/science-and-technology/review-erik-larson-myth-of-artificial-intelligence/
 

Thread Starter

Pushkar1

Joined Apr 5, 2021
401
today is just a different way to program in the standard way.
I am curious to know what it means to train a model I don't understand what it is to train the model in the program.

I have read in many articles that model has to be trained but I don't understand here what it means to train a model because once the system is programmed. it will do exactly what it's supposed to instruct.

For example if I take a room cleaning model, what does it mean to train it?
 

xox

Joined Sep 8, 2017
608
I am curious to know what it means to train a model I don't understand what it is to train the model in the program.


I have read in many articles that model has to be trained but I don't understand here what it means to train a model because once the system is programmed. it will do exactly what it's supposed to instruct.


For example if I take a room cleaning model, what does it mean to train it?
Think of it this way. You are trying to approximate a function which has many inputs and outputs. The weights and biases in a neural network (NN) for example can be viewed as the coefficients of a very large polynomial. Training that network then is simply the act of feeding input into it, comparing the output with the "expected" result, then adjusting the coefficients accordingly. For the traditional NN an algorithm called "gradient descent" is used to calculate just how much each weight and bias should be adjusted.

The training step isn't however so straightforward. You can end up with an undertrained model if incomplete/inaccurate data is used, one that is overtrained if too many inputs of a "certain quality" are fed into it, or even one that is easily "fooled" by particular combinations of input parameters due to some sort of "unfortunate coincidence". So the training aspect could almost be considered something of a science in and of itself. Just something to keep in mind.

If I were you I'd start small. Build a traditional multi-perceptron-based NN that takes an arbitrary number of inputs and maps them to an arbitrary number of outputs. Multiple "hidden" layers just complicate things and anyway they aren't necessary so just one will suffice. Also, the XOR function is a good first choice for training because it isn't linearly separable. (And thus easily verified.)
 
Top