Turbulent odour plumes are challenging to track since there is usually no smooth gradient leading to the source. Instead, it was thought that insects used odour encounters and wind direction as main clues for foraging or locating mates. Recently, it was described that fruit flies can also use “stereo”-information inferred from temporal correlations between their two antennas [1].

Here we validate this strategy using robotic experiments in a turbulent environment.

We had previously constructed a stereo artificial nose based on metal-oxide sensors [2] which could resolve the onset of odorants with sub-second accuracy, inferring odour direction from temporal correlations similar to fruit flies.

We equipped a wheeled robot with a pair of sensors and placed it in a room with a fan and an intermittent odour source. Sensor output spikes fed into a Time Difference Encoder (TDE) [3], a neural model encoding the time difference between two events into the number of spikes contained in an output burst.

A neural network comprising onset filters, TDEs and cross inhibition was able to perform lateralization of the gas source. A reactive navigation algorithm directed the robot towards the side of first detection and finally to the source, supporting the concept of stereo olfaction for plume navigation.

Further work could evaluate, e.g., the optimal sensor separation distance and use wind direction as an additional clue. Using stereo olfaction to bias a random walk could allow the robot to start outside the odour plume.

Funding
DD, MS: EU H2020 #785907, #945539 (HBP)
MS: MRC #MR/T046759/1, NSF #2014217 (NeuroNex Odor2Action)
TS, EC: CogniGron, U. Emmius Funds (U Groningen). Exc. Cluster 227 CITEC U Bielefeld
TS: CapoCaccia ‘19 NEUROTECH fellowship

References
[1] Kadakia et al. (2021) https://doi.org/10.1101/2021.09.29.462473
[2] Drix & Schmuker (2021) https://doi.org/10.1021/acssensors.0c02006
[3] Gutierrez-Galan et al. (2021) https://doi.org/10.1109/TNNLS.2021.3108047