AI (Artificial intelligence) ...

[marc_256]

Hi,

First at all, I just love the DARPA Robotics Challenge ...

For my new (latest) robot,
I like to build my first humanoid robot using AI (Artificial intelligence).
So i will need a lot of Algorithms.

I purchased the LATHE PANDA DELTA with 4GB Ram, 32GB Flash memory and 500GB SSD.
This SBC (Single board computer) have full window 10 pro x64 on board,
and an arduino leonardo who haves a lot of user I/O's and communication channels.

Is here someone who did some stuff with AI using PB ?

thanks,
Marco

[Cyllceaux]

Long time ago I wrote a agent-system in PB. It's more like the Blueprintsystem from UE5, but Agents are a sub-category of AI. So, the question is... what do you mean with AI.
- Agents
- Neural Networks
- Cell replications
- Pattern recognition
- or other stuff like big "if-then-else" blocks

It's a really big field.

[marc_256]

Hello Cyllceaux,

thanks for your answer,

Well, the challenge of the robot contest is to find 5 correct cans on the floor.
My DIY robot have now a DIY robot arm with pneumatic gripper, and can move around.
For the robot arm control I use IK (Inverse Kinematics)

See simulator here:
https://www.purebasic.fr/english/viewt ... p?t=80493

They (the cans) are placed on a variable position on the floor
and we need to collect them and breng them to one place (corner).

So, I purchased 2 VGA cameras (640x480 px) who are connected as stereo visions to a teensy 4.1 board.
I am working on a software to store the 2 images (left/right) in RAM memory of the teensy 4.1.
Sended to the LATHE PANDA DELTA memory.


And then comes the trouble, haha

I like to find via these 2 images and some algorithms:

- can position (robot to can)
- distance (robot to can)
- can color detection (to find the correct cans)

I need to select only the red cans, and not the other ones ...

I am very very new to AI, so I think I need some neural networks and Pattern recognition ?
Filtering stored data (can data) in memory with the images in memory.
The distances, I think I can use geometry for that ...

Thanks,
Marco

Image of the robot arm, so far.

[Cyllceaux]

Hi Marco,

To be honest. Every task can be done by an AI. But to 99,9999% you don't need an AI.
I think I describe the Agent-System. Maybe you find an idea (or solution) for your problem.

The idea behind an agent is, you have unlimited agents. every agent did only one task. It has an Input-Signal(s), WorkingUnit, Export-Signal(s)

e.g.

• Agent A makes a picture from the cam and calls the next Agent with a kind of event (in PB like PostEvent)
• Agent B takes the picture and checks if it find a cup and fires the next event. (pictures recognition does not need any kind of AI. It has to check color, form and size. there are some threads in the forum for doing this)
• Agent C get the information of the cup (size, color, position (x,y,z)) and calculates distances. and fires the next event
• Agent D controls the arms
• Agent E checks everything

So you can concentrate on each agent without disturbing others. At the dissertation of an employee, he had an laser as input-signal (instead of a cam) and controlled different automations and logging-systems.

But, this sound really interesting for me. I wrote an controller for our radiators with a raspberrypi and 4 sensors. But no robot.

[Caronte3D]

I think you can do what you need with OpenCV

The user JHPJHP was do a nice work translating it to PureBasic:
(He asks for a donation to get it, but it's worth it)
viewtopic.php?t=57457