Bien sûr! Je serais ravi de vous aider à écrire un résumé journalistique en français. Cependant, vous n’avez pas fourni la transcription de la vidéo YouTube. Pourriez-vous, s’il vous plaît, partager le texte que vous aimeriez que je résume ? Merci !
Source : Nikodem Bartnik | Date : 2024-10-11 18:00:34 | Durée : 00:15:24
➡️ Accéder à CHAT GPT en cliquant dessus
Abonnez vous a Chat GPT
Comments
The first 500 people to use my link https://skl.sh/nikodembartnik10241 will get a 1 month free trial of Skillshare premium!
Check out the second part: https://youtu.be/JXjkAZ5dZJU
It is a really great project! I planned to do something similar with a flying drone. From the video, I can see that the AI has problems calculating distance using only the image, which is why it tends to go directly into obstacles. I think distance sensors could help here.
Also, another issue is that the AI does not remember its main task—to get as close as possible to the rocket. I think you should develop something that behaves like an agent, which in every loop reminds itself of the main goal.
Good luck! It’s really a great project!
I was thinking: why not add some proximity sensors to all sides and let the program translated the obtained measurements into text, which can be added to your prompt. By also providing limits to as far it may move towards an object, I think the motion control by LLM could became more useful.
Impressive. This almost seems like a prototype for an assistant robot for a non-seeing person.
Does the robot know how to estimate the distance between himself and the object he sees?
Would super imposing some markers into the images for navigation help? Then tell the prompt what those markers mean. Hard to explain. For example if the opening is over marker 3, the AI will know it needs to turn 20 degrees left or something.
Mujaheddin change yes Bahutu Achaea lga
can i use esp32?
Robo dexo 2000 ..
I guess u dont has submitted the context of the conversation? U only submitted the "System" and the "User" Message as a new block.
We should be scared
Did you set it priorities as survive at all costs or explore at all costs? It's going to be very entertaining when this thing gets loose from your house and wanders into someplace its really not allowed to go. 😁🤖
Use sensors for safety and perfect maneuver if you can
You just might be the first man in the world to do this.
age of Ultron
Polska Gurom
Nice project!! I am sure you could get better results with better prompts . Your fault was that you didnt use the exported text after sending the image in a new prompt. U need to make a feedback loop inside your prompting. Send me a message i am sure i can help you.
add reasoning logic to the code.
bro just didn't code in the whole video.
Video's Definitely worth subscribing!!
There are courses and papers on MemGPT which is about storing memory within the processes you’re trying to do as a way to use LLMs as central operating systems for agents and/or autonomous systems, as in your case
get bigger body chat
for the path issue,code it in a way so that if the object is at a particular distance,it should avoid it and find a better path
Bro is one of the people that is going to be terminated, by someone from the future cause, he caused the apocalypse.
✅✅ Can you tell me what's the name of that Blue carpet? ✅✅
I built a ChatGPT robot using an iPhone and LEGO, with similar results. I also tried using the TrueDepth camera on the iPhone to capture a 3D depth map which I turned into a 2D map. Eventually the problem boiled down to trying to create an accurate representation of where the robot is in the world (SLAM) which I could not get to run robustly with the iPhone camera’s limited FOV. I also tried using YOLO to create a symbolic representation of the scene, but that was extremely inaccurate. I tried many other things, such as using a bag of words type of vector to try locate the robot from a library of previously seen images – again it was simply not robust enough and the robot’s location was ambiguous. I think I need cameras with wider FOV to see orthogonal as well as parallel to the direction of travel, as well as a stereo pair or LiDAR to capture a robust depth map to detect obstacles ahead.
ask it to make a virtual simulation of your house 🙂
now give it a knife
This is so cool.
Please, Please, Please. Revisit this, but with Deepseek R1
1:42😁
What if the tires had sensors or something similar to know if there are two contacts on each tire? (one on the floor and one that would not be on the floor) It's just an idea that occurs to me, I really don't know about robotics.
U all ready did it
haha thats awesome! as someone who works with AI everyday, im curious about how you handled the physical movement controls? most LLMs like gpt4 are great at processing language but need additional frameworks for robotics. did you use any specific libraries or frameworks to bridge that gap?
you should've put ultra sonic or any other proximity sensor to determine the the distance between the object for more accurate navigation
Skynets real origin
i wonder how we can make AI remember stuff. In OpenAi at least the conversation is remembered up to a point
in 200 years when the AI dominates the human: "I put a human youtuber on a robot and let him explore the world"
świetne ! pozdrawiam ! 😀
why not fix a 360 Cam ?
It kind of cool and at the same time worthless in the aspect of thinking it will destroy or do anything to human when it basically doesn’t av arm that ChatGPT can control