Distance measurement alone will require three points of reference to trilaterate position. I did a machine learning experiment a few years ago to simulate a radio network to see if I could reliably get accurate positions using gradient descent, randomly-positioned nodes, and noisy latency measurements. The result was underwhelming and inaccurate in most cases, though it could perform well under some limited circumstances, e.g. when the known points had a good distribution and the initial guess was not near an incorrect local gradient minimum. However, the computational cost was too high for the purpose I had in mind, so I dropped my efforts. Maybe I'll do some experimentation using other approximation methods or tricks once I get some ESP32s to play with, but I am pretty sure that spanning trees will provide a better solution for greedy routing.
Dear Friend @k98killer *Thank you very much* for your interest in our channel. *_Q_* - _Distance measurement alone will require three points of reference to trilaterate position. I did a machine learning experiment a few years ago......._ ➡️ Yes!, You're absolutely right about the challenges of distance measurement and trilateration! Your machine learning experiment sounds fascinating, and We're not surprised that you faced difficulties with accuracy and computational cost. Trilateration can be tricky, especially with noisy measurements and limited reference points. Gradient descent can get stuck in local minima, leading to inaccurate results. Spanning trees are a great approach for greedy routing, as they can efficiently connect nodes in a network while minimizing the total edge weight (e.g., distance or latency). If you're interested in exploring alternative methods, you might consider: 1. Multilateration: Using four or more reference points to improve accuracy. 2. Non-linear least squares: A optimization technique that can handle noisy measurements. 3. Graph-based methods: Like graph neural networks or graph-based SLAM (Simultaneous Localization and Mapping). 4. Hybrid approaches: Combining trilateration with other techniques, like inertial measurement units (IMUs) or vision-based methods. 📌 *Thank you very much* for sharing information and experience about your experiments and ideas! Feel free to share, we're also excited to see your progress. This is a very nice project you are working on. Happy Tinkering. 👍 If you have any queries/issues/suggestions about this Tutorial, or any of our DIY project, feel free to ping us on WhatsApp/Telegram (Links given below) for further support. Tech support - (𝗖𝗵𝗮𝘁 𝗢𝗻𝗹𝘆) Telegram/WhatsApp - 11:00 AM - 05:00 PM (Mon-Fri) IST 🚀 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺 🔗 t.me/make2explore 💬 𝗪𝗵𝗮𝘁𝘀𝗔𝗽𝗽 🔗 bit.ly/3VXGGEe 📩 info@make2explore.com / support@make2explore.com Best Regards
![[Device Overview] Adafruit Circuit Playground Express](http://i.ytimg.com/vi/tHRV9FFY870/mqdefault.jpg)
![[Device Overview] Adafruit Circuit Playground Express](/img/tr.png)
![[Device Overview] Ultra-wideband Transceiver Module RYUW122](http://i.ytimg.com/vi/HwBakNIPAz4/mqdefault.jpg)
Distance measurement alone will require three points of reference to trilaterate position. I did a machine learning experiment a few years ago to simulate a radio network to see if I could reliably get accurate positions using gradient descent, randomly-positioned nodes, and noisy latency measurements. The result was underwhelming and inaccurate in most cases, though it could perform well under some limited circumstances, e.g. when the known points had a good distribution and the initial guess was not near an incorrect local gradient minimum. However, the computational cost was too high for the purpose I had in mind, so I dropped my efforts. Maybe I'll do some experimentation using other approximation methods or tricks once I get some ESP32s to play with, but I am pretty sure that spanning trees will provide a better solution for greedy routing.
Dear Friend @k98killer
*Thank you very much* for your interest in our channel.
*_Q_* - _Distance measurement alone will require three points of reference to trilaterate position. I did a machine learning experiment a few years ago......._
➡️ Yes!, You're absolutely right about the challenges of distance measurement and trilateration! Your machine learning experiment sounds fascinating, and We're not surprised that you faced difficulties with accuracy and computational cost.
Trilateration can be tricky, especially with noisy measurements and limited reference points. Gradient descent can get stuck in local minima, leading to inaccurate results.
Spanning trees are a great approach for greedy routing, as they can efficiently connect nodes in a network while minimizing the total edge weight (e.g., distance or latency).
If you're interested in exploring alternative methods, you might consider:
1. Multilateration: Using four or more reference points to improve accuracy.
2. Non-linear least squares: A optimization technique that can handle noisy measurements.
3. Graph-based methods: Like graph neural networks or graph-based SLAM (Simultaneous Localization and Mapping).
4. Hybrid approaches: Combining trilateration with other techniques, like inertial measurement units (IMUs) or vision-based methods.
📌 *Thank you very much* for sharing information and experience about your experiments and ideas! Feel free to share, we're also excited to see your progress. This is a very nice project you are working on. Happy Tinkering. 👍
If you have any queries/issues/suggestions about this Tutorial, or any of our DIY project, feel free to ping us on WhatsApp/Telegram (Links given below) for further support.
Tech support - (𝗖𝗵𝗮𝘁 𝗢𝗻𝗹𝘆) Telegram/WhatsApp - 11:00 AM - 05:00 PM (Mon-Fri) IST
🚀 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺 🔗 t.me/make2explore
💬 𝗪𝗵𝗮𝘁𝘀𝗔𝗽𝗽 🔗 bit.ly/3VXGGEe
📩 info@make2explore.com / support@make2explore.com
Best Regards