Okay I keep getting mixed answers. Vex AI is not actuality AI it is all autonomous coding right? And do you use VRC parts or Vex U parts or what. Sorry if the answers are right in front of me. Thanks for helping me understand.
VEX AI uses AI developed by VEX. Students don’t actually have to mess with it and can just treat the provided AI kit as one big sensor for detecting objects & robot positions, just focusing on autonomous coding.
For my team, I do want to mess with the Jetson Nano (Linux board that does the AI processing for the depth camera and fancy AI neural-network camera we’re provided) and get it to detect objects using our own AI code rather than VEX’s AI at some point, but I’m not super familiar with computer vision so it’ll probably be difficult.
As for parts, the rules are similar to VEX U. Here’s a snippet from the Change Up manual’s AI appendix explaining what we’re allowed to use:
10 Likes
I think there is a issue of nomenclature here. What do you mean by “its not actually AI”.
7 Likes
The AI part of VEX AI is VEX using a buzzword to hype up the new competition.
The new competition has robots that compete fully autonomously. One way to control these robots is using AI.
5 Likes
I mean like it can’t think by itself.
Okay so it has sensors and is solely based on programming. Thanks guys
1 Like
The AI part of VEX AI is VEX using a buzzword to hype up the new competition.
Computer vision could be considered a form of AI, and VEX is using a sensor that uses a deep neural network internally to detect the field code. Neural networks are (kinda) modeled after the human brain, so to me, that sensor can have a model that can “think”/“learn” loaded onto it. But I guess it depends on your definition of “think by itself.”
AI for robot control probably will not be necessary, I think a PID loop would probably work just fine for controlling the robot, especially since the robot knows its own position as it moves and can have any error introduced by wheel slipping removed by the absolute reading of the GPS sensor. I haven’t heard of any VRC or U team that uses AI for motor control, as there are other algorithms they use for precise motion like Pathfinder or Pure Pursuit.
4 Likes
Ya so my guess what you are referring to is the holy grail of AI. AGI, artificial general intelligence. It doesn’t exist, depending on what AI researcher you ask its between 10 and 100 years away.
This point is a bit more complicated. But the generally accepted answer is most of those tasks are not AI. They are robot control, or motion planning tasks.
(There is some disagreement if computer vision is a subset of AI or its own field.)
However the 1 part of a fully autonomous scoring VEX robot that is 100% definitely AI is the strategy and decision making part. The task VEX left completely up to the teams in the VEX AI competition is purely AI, where you are an agent playing a game. When someone says they do research in AI, this is the area they are most likely in. This matches with the simple “AI” you will see in a video game playing against you, they are strategy/decision making agents albeit not very sophisticated ones.
13 Likes
Following up, this is a video of the AI work I did for the fully autonomous robot I was working on for ITZ. You can see I ran the AI code in a simplified model of the world where I could search for strategies much more efficiently.
I like to think of these AI’s as like a drive coach, they should tell you high level things and trust you know how to drive your robot.
- The number on a tile is how many cones are there. Past “get cone in that tile” its not really worth planning at a high level.
- The bottom left 4 tiles are where you can place mobile goals for points for red, top right for blue.
- The red 13 and blue 13 are cones that only those color robots can pick up. They are from the loading station but the only important detail for the AI is that they are red exclusive cones.
- The big circles are mobile goals, the number on them is cones stacked on top.
The cool thing about building a little game engine like this is I can both test different AI’s against each other and simulate games if I want to do a machine learning approach.
18 Likes
Wikipedia has one definition of AI as:
Leading AI textbooks define the field as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[
so I think a robot that understands its location, where the other robots on the field are and what objects may be in its way, perceives its environment and can be considered using AI to achieve its goal.
13 Likes
Thanks so much that really helps!
1 Like
This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.