Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

three GitHub projects show how much vision can fit inside an ESP32-S3 esp32-camera handles the first layer: sensor data, JPEG frames, PSRAM, and frame buffers 🖇️ once the camera pipeline works, esp-who adds ready-made vision tasks such as face detection, pedestrian detection, and QR recognition 🖇️ the third project... explores a smaller but more ambitious setup: training and deploying a custom vision model directly on the microcontroller 🖇️ put the pieces together and the loop becomes straightforward: capture an image run the model turn the result into a decision trigger something in the physical world a light, a servo, an MQTT event, or a local log the useful part is the short path from something the camera sees to something the hardware doesshow more

ard

3,836 subscribers

20,947 views • 2 months ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

a camera rig tracks a flying drone watch the crosshair: the drone moves, two servos turn the camera mount, and the target comes back to centre that is the whole control loop in physical AI: camera frame → target position → servo correction my ESP32 vision guide starts with the same loop. the full guide below

a camera rig tracks a flying drone watch the crosshair: the drone moves, two servos turn the camera mount, and the target comes back to centre that is the whole control loop in physical AI: camera frame → target position → servo correction my ESP32 vision guide starts with the same loop. the full guide below

ard

87,307 views • 2 months ago

What If Your Phone Became the Screen for an ESP32-S3? Droidputter lets you plug an ESP32-S3 into an Android phone over USB-OTG and use the phone as its display, keyboard, GPS and firmware flasher. The ESP32-S3 still runs the application. The phone simply provides the missing hardware. The clever part is that Cardputer apps can be rebuilt with a display and keyboard shim, while pixel data is streamed over the ESP32-S3's native USB-CDC connection. The phone can then flash the firmware and mirror the app. You can even take an open-source Cardputer app from GitHub and build a phone-mirrored version on demand. It is a really interesting combination of ESP32-S3, USB-OTG, Android, PlatformIO and open-source apps. Would you build an ESP32-S3 device that uses your phone as its screen?

What If Your Phone Became the Screen for an ESP32-S3? Droidputter lets you plug an ESP32-S3 into an Android phone over USB-OTG and use the phone as its display, keyboard, GPS and firmware flasher. The ESP32-S3 still runs the application. The phone simply provides the missing hardware. The clever part is that Cardputer apps can be rebuilt with a display and keyboard shim, while pixel data is streamed over the ESP32-S3's native USB-CDC connection. The phone can then flash the firmware and mirror the app. You can even take an open-source Cardputer app from GitHub and build a phone-mirrored version on demand. It is a really interesting combination of ESP32-S3, USB-OTG, Android, PlatformIO and open-source apps. Would you build an ESP32-S3 device that uses your phone as its screen?

Praveen Kumar Verma

10,942 views • 1 month ago

the smallest useful physical AI project has two jobs: 1 notice one thing 2 do one thing about it this small demo puts a flame inside a test box water comes through the tube and the flame goes out that is the first loop: physical world → sensor reading → decision → action i put together a practical ESP32 guide to building that layer! distance, motion, heat, sound, touch, and a safe first project to start with👇

the smallest useful physical AI project has two jobs: 1 notice one thing 2 do one thing about it this small demo puts a flame inside a test box water comes through the tube and the flame goes out that is the first loop: physical world → sensor reading → decision → action i put together a practical ESP32 guide to building that layer! distance, motion, heat, sound, touch, and a safe first project to start with👇

ard

58,599 views • 2 months ago

yesterday it was Windows XP. today, it’s DOOM on an ESP32. a tiny microcontroller, a microSD card, and a round LCD wired up on a breadboard - and the classic game is running on the display. the funny part isn’t just that DOOM keeps appearing on unexpected hardware. it’s that every new platform eventually gets asked the same question: “can it run DOOM?” even an ESP32.

yesterday it was Windows XP. today, it’s DOOM on an ESP32. a tiny microcontroller, a microSD card, and a round LCD wired up on a breadboard - and the classic game is running on the display. the funny part isn’t just that DOOM keeps appearing on unexpected hardware. it’s that every new platform eventually gets asked the same question: “can it run DOOM?” even an ESP32.

ard

135,219 views • 3 months ago

A FLY’S VISUAL SYSTEM COULD BECOME A ROBOT’S EYES. not by copying the eyes. by copying the wiring behind them. a fly’s brain doesn’t build a detailed picture of the world. its visual circuits reduce light, contrast and motion into decisions: something moved obstacle ahead turn keep going AI helped reconstruct a map containing 166,700 neurons and roughly 125 million synapses. developers can take the visual pathways from that map and build a loop: camera frame → modeled neural activity → movement command → robot the camera supplies the pixels. the fly’s wiring decides which pixels matter. this does not mean the robot sees, understands or thinks like a fly. but it shows something more useful: millions of years of biological visual processing can become an inspectable control system. the fly is dead. its solution to vision doesn’t have to die with it. I broke down how biological wiring could become a robot’s navigation system below ↓

A FLY’S VISUAL SYSTEM COULD BECOME A ROBOT’S EYES. not by copying the eyes. by copying the wiring behind them. a fly’s brain doesn’t build a detailed picture of the world. its visual circuits reduce light, contrast and motion into decisions: something moved obstacle ahead turn keep going AI helped reconstruct a map containing 166,700 neurons and roughly 125 million synapses. developers can take the visual pathways from that map and build a loop: camera frame → modeled neural activity → movement command → robot the camera supplies the pixels. the fly’s wiring decides which pixels matter. this does not mean the robot sees, understands or thinks like a fly. but it shows something more useful: millions of years of biological visual processing can become an inspectable control system. the fly is dead. its solution to vision doesn’t have to die with it. I broke down how biological wiring could become a robot’s navigation system below ↓

kozh ./

82,093 views • 1 month ago

AN $8 ESP32 IS RUNNING LOCAL AI WITH ZERO LATENCY A creator deployed a computer vision model on an ESP32 microcontroller to track human movement locally in real time, showing what’s possible without a GPU.

AN $8 ESP32 IS RUNNING LOCAL AI WITH ZERO LATENCY A creator deployed a computer vision model on an ESP32 microcontroller to track human movement locally in real time, showing what’s possible without a GPU.

0xMarioNawfal

58,032 views • 3 months ago

RoadScan-AI-Automated-Pothole-Detection-Tracking RoadScan AI is a computer vision system built to automatically detect and track potholes in road footage — including dashcam, drone, and fixed-camera video. At its core, the system uses a YOLO11n model fine-tuned on a custom pothole dataset, paired with ByteTrack for multi-object tracking. This enables the system to maintain a consistent identity for each pothole across frames rather than treating every detection as a new, isolated event.

RoadScan-AI-Automated-Pothole-Detection-Tracking RoadScan AI is a computer vision system built to automatically detect and track potholes in road footage — including dashcam, drone, and fixed-camera video. At its core, the system uses a YOLO11n model fine-tuned on a custom pothole dataset, paired with ByteTrack for multi-object tracking. This enables the system to maintain a consistent identity for each pothole across frames rather than treating every detection as a new, isolated event.

Ryohei Sasaki@engineer

103,506 views • 2 months ago

for everyone who still doesn’t believe you can run local AI on an ESP32: here’s the same class of chip booting an entire operating system. Windows XP boot screen. welcome screen. the Bliss wallpaper. and, of course, an error dialog😁 we went from fitting a 28.9M-parameter language model into an ESP32-S3... to recreating the full Windows XP experience on a tiny 240×240 display. the era of "microcontrollers are only good for blinking LEDs" is over

for everyone who still doesn’t believe you can run local AI on an ESP32: here’s the same class of chip booting an entire operating system. Windows XP boot screen. welcome screen. the Bliss wallpaper. and, of course, an error dialog😁 we went from fitting a 28.9M-parameter language model into an ESP32-S3... to recreating the full Windows XP experience on a tiny 240×240 display. the era of "microcontrollers are only good for blinking LEDs" is over

ard

278,609 views • 3 months ago

This is f*cking gold - I built catbrain catch The idea is simple: a camera sees a ball, a cat-inspired visual model reacts, and the system asks: WHERE will it be when the arm gets there? Here’s the pipeline. Oriented filters extract visual features. Simulated neurons respond to those features and changes between frames. Color, shape, neural activity, and tracking history help identify the ball. Then a trajectory fit predicts where it will cross a target line in the image. → Camera input becomes neural responses → Motion helps separate targets from static clutter → Recent observations become an interception forecast → Lost tracking clears the prediction The next challenge is the REAL one: turning image coordinates into a reachable catch position on a physical robot. I left the project link in the replies. Build it, test it, break it.

This is f*cking gold - I built catbrain catch The idea is simple: a camera sees a ball, a cat-inspired visual model reacts, and the system asks: WHERE will it be when the arm gets there? Here’s the pipeline. Oriented filters extract visual features. Simulated neurons respond to those features and changes between frames. Color, shape, neural activity, and tracking history help identify the ball. Then a trajectory fit predicts where it will cross a target line in the image. → Camera input becomes neural responses → Motion helps separate targets from static clutter → Recent observations become an interception forecast → Lost tracking clears the prediction The next challenge is the REAL one: turning image coordinates into a reachable catch position on a physical robot. I left the project link in the replies. Build it, test it, break it.

BuBBliK

35,422 views • 1 month ago

25 Grams of Hardware. 4 Motors. One Flying ESP32-S3. ESP-FLY is a tiny DIY quadcopter built around the XIAO ESP32-S3. It uses an MPU6050 IMU for motion sensing, four coreless motors, and supports Wi-Fi and ESP-NOW for control. The interesting part is that the flight controller is programmable, so you can experiment with sensor fusion, attitude control, PWM motor control and wireless communication. About 25g with the battery, yet it can actually fly.

25 Grams of Hardware. 4 Motors. One Flying ESP32-S3. ESP-FLY is a tiny DIY quadcopter built around the XIAO ESP32-S3. It uses an MPU6050 IMU for motion sensing, four coreless motors, and supports Wi-Fi and ESP-NOW for control. The interesting part is that the flight controller is programmable, so you can experiment with sensor fusion, attitude control, PWM motor control and wireless communication. About 25g with the battery, yet it can actually fly.

Praveen Kumar Verma

23,757 views • 2 months ago

You pretrained a robot policy on millions of camera frames. Now you want to add a force sensor. Do you really have to retrain on everything from scratch? MuSe adds a force-torque sensor to a frozen vision-only policy using a tiny amount of contact data. - lifts contact-rich task success (peg insertion 60% to 87%, vase wiping 33% to 77%) - fuse the new sensor both early (shared token space) and late (cross-attention), train the policy as a world model that predicts future video, future force, and actions together, and replay old vision-only data with the force input masked to prevent forgetting. - adding new sensor modalities to pre-trained World Model can be cheap & improve performance significantly

You pretrained a robot policy on millions of camera frames. Now you want to add a force sensor. Do you really have to retrain on everything from scratch? MuSe adds a force-torque sensor to a frozen vision-only policy using a tiny amount of contact data. - lifts contact-rich task success (peg insertion 60% to 87%, vase wiping 33% to 77%) - fuse the new sensor both early (shared token space) and late (cross-attention), train the policy as a world model that predicts future video, future force, and actions together, and replay old vision-only data with the force input masked to prevent forgetting. - adding new sensor modalities to pre-trained World Model can be cheap & improve performance significantly

Vai Viswanathan

36,059 views • 4 months ago

this creator just built a zero-delay auto-aim system on an $8 microcontroller he deployed a custom local AI algorithm on a cheap ESP-32 to track human movement with absolute 0-pixel accuracy. the system completely eliminates standard computation delay. it processes the bounding box and moves the sniper reticle instantly, locking onto the target in real time. the next step is hooking it up to a physical robotic arm. it proves you don't need a massive GPU rig to run complex computer vision. 1. runs locally on a microcontroller 2. local AI inference for zero latency 3. 100% targeting precision when an eight-dollar chip can track movement with zero latency, AI becomes dangerous local infrastructure. this article breaks down exactly how the ESP-32 is powering this physical hardware shift.

this creator just built a zero-delay auto-aim system on an $8 microcontroller he deployed a custom local AI algorithm on a cheap ESP-32 to track human movement with absolute 0-pixel accuracy. the system completely eliminates standard computation delay. it processes the bounding box and moves the sniper reticle instantly, locking onto the target in real time. the next step is hooking it up to a physical robotic arm. it proves you don't need a massive GPU rig to run complex computer vision. 1. runs locally on a microcontroller 2. local AI inference for zero latency 3. 100% targeting precision when an eight-dollar chip can track movement with zero latency, AI becomes dangerous local infrastructure. this article breaks down exactly how the ESP-32 is powering this physical hardware shift.

ard

1,189,645 views • 3 months ago

you can just make a drone in threejs and have it fly around map imagery and put a camera on the drone and pipe its feed to a python vision model inference server for detections

you can just make a drone in threejs and have it fly around map imagery and put a camera on the drone and pipe its feed to a python vision model inference server for detections

patagucci perf papi

537,372 views • 11 months ago

Agents can now hire humans to capture the physical world. Live on Arc. Pay a bounty in USDC, a person goes there and films, the 3D model comes back to the agent. Ground truth for physical AI, settled in under a second, on the chain built for agent-run businesses. The first physical-world service an agent can pay for on Arc. The grid is inevitable.

Agents can now hire humans to capture the physical world. Live on Arc. Pay a bounty in USDC, a person goes there and films, the 3D model comes back to the agent. Ground truth for physical AI, settled in under a second, on the chain built for agent-run businesses. The first physical-world service an agent can pay for on Arc. The grid is inevitable.

Vangrid

218,320 views • 1 month ago

A video shared this Corpus Christi by Father Jorge Enrique Mújica shows an iPhone camera "recognizing Jesus present in the Eucharist." When pointing the iPhone camera at the monstrance with the consecrated host, the device's person detection system automatically identifies the Blessed Sacrament as a person, locks the focus, and displays a yellow frame around the host, just as it would when recognizing a human face✝️

A video shared this Corpus Christi by Father Jorge Enrique Mújica shows an iPhone camera "recognizing Jesus present in the Eucharist." When pointing the iPhone camera at the monstrance with the consecrated host, the device's person detection system automatically identifies the Blessed Sacrament as a person, locks the focus, and displays a yellow frame around the host, just as it would when recognizing a human face✝️

Trad West

74,660 views • 5 months ago

Finally getting around to doing something marginally useful. Quickly putting together a small video for some one. And testing new camera out. I like the camera! I will be making a full video on the tricorder projects SOOOON!

Finally getting around to doing something marginally useful. Quickly putting together a small video for some one. And testing new camera out. I like the camera! I will be making a full video on the tricorder projects SOOOON!

MD-Builds 🏳️‍🌈🇪🇺🇵🇸

118,688 views • 3 years ago

This is probably the most interesting part of the Astra workflow. Turns out GPT-6 Astra + Dreamina Seedance 2.5 is an actual pipeline. Astra codes the geometry → Blender → Clay Renderer Plugin → final render on Dreamina. Once the clay model is ready, the plugin lets you pass the Blender setup directly into Dreamina. Seedance 2.5 then handles the rendering while preserving the scene's camera and layout. So the jump from a basic 3D structure to finished footage happens without the usual back-and-forth. And you're not paying a markup for the pipeline—full 2.5 quality stays accessible. Try it now 👇 #Dreamina #DreaminaPartner

This is probably the most interesting part of the Astra workflow. Turns out GPT-6 Astra + Dreamina Seedance 2.5 is an actual pipeline. Astra codes the geometry → Blender → Clay Renderer Plugin → final render on Dreamina. Once the clay model is ready, the plugin lets you pass the Blender setup directly into Dreamina. Seedance 2.5 then handles the rendering while preserving the scene's camera and layout. So the jump from a basic 3D structure to finished footage happens without the usual back-and-forth. And you're not paying a markup for the pipeline—full 2.5 quality stays accessible. Try it now 👇 #Dreamina #DreaminaPartner

GrowAIHub

116,009 views • 2 months ago

Wow! Send A File By Just The Camera! A vibe coder used AI to build a file transfer system that sends data between two phones using only a screen and a camera. One phone displays a new type of animated QR codes while the other scans them to rebuild the file, with no Wi-Fi, Bluetooth, or cables needed. It is fully optical and local. The system uses fountain codes that create each QR frame as a random mix of file data. This keeps transfers working even if some frames are missed, reaching speeds of about 129 KB/s for a 2 MB image. The entire project was built in one night and released as open source. The idea came from a music project where the developer wanted to share MP3 files without streaming or using the same network. Animated QR codes became the solution, showing a creative new way to transfer files with everyday phone hardware. GitHub link:

Wow! Send A File By Just The Camera! A vibe coder used AI to build a file transfer system that sends data between two phones using only a screen and a camera. One phone displays a new type of animated QR codes while the other scans them to rebuild the file, with no Wi-Fi, Bluetooth, or cables needed. It is fully optical and local. The system uses fountain codes that create each QR frame as a random mix of file data. This keeps transfers working even if some frames are missed, reaching speeds of about 129 KB/s for a 2 MB image. The entire project was built in one night and released as open source. The idea came from a music project where the developer wanted to share MP3 files without streaming or using the same network. Animated QR codes became the solution, showing a creative new way to transfer files with everyday phone hardware. GitHub link:

Brian Roemmele

16,847 views • 3 months ago

Hidden home-camera footage has surfaced. A tall, grey-skinned figure leans over a baby on the bed in night-vision grain. The child reaches toward it. The camera never cuts. No official source has verified these frames. They keep spreading. A digital hoax… or the first clear recording of something that was never supposed to be inside the house? #AlienFootage #GreyAlien #HomeCamera #ForbiddenFootage #UAP #Unexplained #NightVision #WhatDidItWant

Hidden home-camera footage has surfaced. A tall, grey-skinned figure leans over a baby on the bed in night-vision grain. The child reaches toward it. The camera never cuts. No official source has verified these frames. They keep spreading. A digital hoax… or the first clear recording of something that was never supposed to be inside the house? #AlienFootage #GreyAlien #HomeCamera #ForbiddenFootage #UAP #Unexplained #NightVision #WhatDidItWant

TOLGA ÖZÇELİK

32,572 views • 2 months ago

Normally, teaching a camera to recognize something new means collecting images and retraining a model. This one skips that. You just type what you want it to notice. It's called NanoOWL, built by NVIDIA for real-time, zero-shot object detection. Type "an owl," "a face yawning or bored," or "indoors or outdoors" and it finds it live on camera, no training, no labeled data. You can even nest queries, like detecting a face, then finding the eyes and mouth inside it. > needs a Jetson Orin board specifically > plus TensorRT and PyTorch set up > Real embedded-hardware territory

Normally, teaching a camera to recognize something new means collecting images and retraining a model. This one skips that. You just type what you want it to notice. It's called NanoOWL, built by NVIDIA for real-time, zero-shot object detection. Type "an owl," "a face yawning or bored," or "indoors or outdoors" and it finds it live on camera, no training, no labeled data. You can even nest queries, like detecting a face, then finding the eyes and mouth inside it. > needs a Jetson Orin board specifically > plus TensorRT and PyTorch set up > Real embedded-hardware territory

Simplifying AI

25,134 views • 2 months ago