How do you track a bumblebee that weighs less than a grain of rice?

The weight problem

Thirty-eight milligrams. That is the total mass of the receiver including its power source. The device is lighter than two grains of rice.

Standard satellite navigation cannot work at this scale. GNSS chips need more power and more board space than a flying insect can carry. The paper addresses this by removing the receiver's burden almost entirely. The heavy lifting moves to the ground.

How the rotating transmitter works

Three hundred metres. That is the range of each rotating high-gain transmitter. The transmitter sweeps its beam across the landscape. The tiny receiver on the bee only measures signal strength — RSS — a handful of times per rotation. No angle hardware. No phase array. Just a power detector.

Fewer measurements mean less energy spent listening. The receiver consumes less than one hundred eighty microwatts while tracking. If the system takes more RSS samples per rotation the accuracy improves to roughly ten metres but the power budget rises to under six hundred microwatts. Still a fraction of a milliwatt. The trade-off is explicit: three times the listening budget buys a third better precision. For a device that weighs thirty-eight milligrams that budget remains negligible.

What the Gaussian process adds

Signal strength alone is noisy. Multipath and foliage bend the simple distance-to-strength relationship. Ground reflections add further distortion. The authors model the bee's trajectory as a Gaussian process. This imposes smoothness: the insect cannot teleport. The path reconstruction uses doubly stochastic variational inference to fuse the sparse RSS snapshots into a continuous estimate.

The result is approximately fifteen metres accuracy across a scalable landscape. The paper demonstrates the system on Bombus terrestris nest return flights. How many bees were tracked and how long each flight lasted — the document does not say.

Where the fifteen-metre figure comes from

Fifteen metres is not a hard ceiling. It is the reported accuracy at the lower power setting with the minimal measurement count. The same hardware can reach ten metres if the receiver wakes more frequently.

The transmitter rotation speed, the exact number of RSS samples per rotation at each setting, and the update rate of the position estimate — none of these appear in the abstract. The paper itself contains them. The announcement stops at the headline numbers.