Bacardo

Product designer @ REA . Geekery inventor @proudlygeeks

[MalvinasData [Hernán] @MalvinasData]

🎵 No lo soñé 🎶

[Brian Roemmele @BrianRoemmele]

Bee-Inspired Navigation Robot Pinpoints Its Home Using a Neural Network

Saving $1000s of dollars on compute by simulation of the Bee.

How a tiny drone learns to fly home like a honeybee without heavy maps or massive computing power
In the world of robotics, navigation has long been a heavyweight affair. Most autonomous drones and ground robots rely on sophisticated systems like SLAM (Simultaneous Localization and Mapping) that build detailed digital maps of their surroundings in real time.

These approaches deliver pinpoint accuracy, but they come at a steep cost in processing power, memory, and energy, making them impractical for truly small, lightweight, or long-endurance robots.

Now, a breakthrough reported in Nature offers a far lighter alternative. In a research paper titled “Efficient robot navigation inspired by honeybee learning flights,” a paper demonstrated a flying robot that uses a bee-inspired strategy to find its way home, even after flights of hundreds of meters, using neural networks small enough to fit in a few kilobytes.

The accompanying News and Views commentary by neuroethologist Barbara Webb of the University of Edinburgh highlights just how elegant and efficient this bio-mimetic approach is compared to conventional robot navigation.

Learning to Fly Home the Bee Way
Honeybees are masters of long-distance navigation. Before embarking on a foraging trip that can span kilometers, young bees perform short “learning flights” or “orientation flights” in a looping, spiral pattern around the hive.

During these brief excursions, they memorize visual snapshots of the surrounding landscape while keeping track of their position relative to home using an internal sense of direction and distance (path integration).

The new robotic system, dubbed Bee-Nav, copies this exact behavior.

1 Learning phase: The drone performs a short exploratory flight around its “home” base. As it flies, an omnidirectional (360 degree) camera captures panoramic images while the robot simultaneously tracks its position using a simple path-integration system (based on optical flow from the camera, a laser rangefinder for height, and a gyroscope for heading).

2 Training phase: These images are paired with the corresponding “home vector” (direction and distance back to base) to train a compact neural network in a self-supervised manner. No external labels or GPS are needed.
3 Mission phase: The drone flies out on a task, potentially hundreds of meters away, updating its position estimate via path integration.

4 Return phase: On the way back, it heads straight toward the estimated home location using path integration. When it enters the small “Learned Homing Area” (LHA) near base (typically just 0.25 to 10 percent of the total area explored), the visual neural network takes over, outputting a corrected home vector that cancels out accumulated drift and guides the drone to within 0.5 meters of home.

The neural networks are astonishingly small: a basic convolutional model uses just 3.4 kilobytes (868 parameters); a more advanced attention-based version uses 42 kilobytes.

Both run comfortably on a low-power Raspberry Pi 4.
Real-World Performance That Impresses
In simulations and physical tests, Bee-Nav proved remarkably robust:

•Indoor arenas (10 by 10 meters and 30 by 40 meters): 100 percent success rate across dozens of flights, even with obstacles. Homing error was typically under 40 degrees in heading and 1.5 meters in distance.

•Outdoor flights (up to 600 meters in windy conditions of 5 to 10 meters per second): 100 percent success for 30 to 110 meter missions and about 70 percent success for 200 to 600 meter flights. The drone still returned to within 0.5 meters of home in most cases.

These results were achieved without GPS, without building any map, and without high-end GPUs, just a tiny drone carrying minimal computing hardware.

1 of 2

[JT @jiratickets]

boomers sending DMs to AI-generated women on instagram

[Ollie Williams @hypeddev]

New CSS text-fit property https://t.co/nlJEF2f0yc

[Andrew Jiang @andrewjiang]

Generative art golf is my favorite corner of X

make the most beautiful digital art possible with the absolute fewest characters of code

pure wizardry

[Massimo @Rainmaker1973]

The Dutch paper engineer who creates foldable memes.

[✂️ @PaperPaul2]

[Mathematica @mathemetica]

max⁡(∣x∣,∣y∣,∣z∣,∣w∣)=1

You’re witnessing a 2D projection of a 3D shadow cast by a 4D tesseract.

As it rotates through the W-axis, the "inner" and "outer" cubes swap roles: a spatial inversion that feels like a glitch only because our biology is trapped in 3-space.
It’s a hauntingly beautiful reminder that our "reality" is often just a lower-dimensional cross-section of a much more complex structure. Perspective is everything.

This is used in high-dimensional data visualization, hypercube topologies in parallel computing networks, and exploring the geometry of extra dimensions in theoretical physics.

[Massimo @Rainmaker1973]

Fynn Jackson is an origami artist known for creating incredibly detailed paper sculptures, often folding expressive faces and complex forms from a single sheet of paper.

[Bacardo @donbacardo]

Maths visually explained, love these. How the external angles always sum 360

[Mathematica @mathemetica]

Harmonious Geometry: The Hirajoshi Wave.

Watch as these gravity-defying spheres trace the hauntingly beautiful paths of the C Hirajoshi scale.

Each ball is tuned to a specific frequency within this traditional Japanese pentatonic scale (C, D, Eb, G, Ab), creating a mesmerizing "Polyrhythmic Pendulum" effect. As the balls oscillate at slightly different speeds, they drift into chaotic patterns before perfectly realigning into a breathtaking visual and auditory climax.

From the sharp, angular bounce to the fluid, sweeping curves of the rainbow trails, this is where physics meets fine art.

Credit: project.jdm

[David Sirota @davidsirota]

Destroying the @InternetArchive's @WayBackMachine would be the equivalent of the burning of the Library of Alexandria - one of the worst losses of knowledge in history.

Media giants are now threatening to do this.

We can't let this happen.

Pass it on.

[cardoso @Cardoso]

Usei IA para transformar cenas em proporção 4:3 para formato widescreen, criando a informação das laterais, mantendo consistência espacial e temporal.

[Mathematica @mathemetica]

Touch a bowl of water to excite a superposition of symmetric surface wave modes with different frequencies.

Sunlight refracts through ripples into caustics below. Absent energy dissipation, these waves would theoretically reconstruct original shape.

[Mathematica @mathemetica]

This is multivariable calculus made tactile:

Sculpt the sand & watch the height function z=f(x,y) render its level curves, gradient field & elevation heatmap in real time.

Living partial derivatives with pure topographic ecstasy.

[danqi @ohhdanqi]

did you just say a pool of text???

this is so satisfying i feel like i just dipped my toes in the water

should i make this public?

[Daniel Beauchamp @pushmatrix]

Huh, so that's why text is called a string