Odor direction sensing with stereo antennae improves simulated odor source localization in realistic turbulent plumes
Fri-P2-072
Presented by: Samuel Sutton
A recent study suggested that Drosophila exploit lagged temporal correlation in odor signals between their antennae for navigation [1], similar to the Reichardt-Detector-model measuring optical flow in the fly’s eye [2]. Here we ask how well this “odor-flow” navigation strategy works in realistic plumes.
Turbulent plumes were recorded in a wind tunnel using Planar Laser-Induced Fluorescence [3]. Recordings cover near-bed (NB) and mid-air (“free-stream”, FS) gas release configurations, since plumes exhibit different behaviors depending on odor release and sensing height.
We simulated Drosophila-inspired virtual agents performing source localisation in these plume recordings. The agents turned and moved depending on the odor signal they received in each timestep. Direction-sensing (DS+) agents turned towards the odor when odor-flow (lagged correlation between antennae) was above a threshold. Otherwise, they followed the DS- strategy: turn and move in upwind direction when odor concentration is above a threshold; or pursue a random walk below threshold.
DS+ agents reached the source in 77% of the trials in NB plumes, and 45% in FS conditions, while DS- achieved 74% success rate in NB and 30% in FS conditions, after optimization of all parameters for each combination.
Our results suggest that odor-flow detection via stereo antennae could be most beneficial in NB conditions, consistent with walking flies, and to a lesser amount in FS conditions. We plan to validate these strategies using robotic agents, with a view toward improving real-world odor source navigation.
References
[1] Kadakia et al. (2021) biorxiv https://doi.org/10.1101/2021.09.29.462473
[2] Haag et al. (2004) PNAS https://doi.org/10.1073/pnas.0407368101
[3] Connor et al. (2018) Exp Fluids https://doi.org/10.1007/s00348-018-2591-3
Funding
NSF Next Gen. Networks for Neuroscience #2014217 (JPC, ACT) & MRC #MR/T046759/1 (MS, SS)
EU H2020 HBP SGA3 #945539 (MS)
Turbulent plumes were recorded in a wind tunnel using Planar Laser-Induced Fluorescence [3]. Recordings cover near-bed (NB) and mid-air (“free-stream”, FS) gas release configurations, since plumes exhibit different behaviors depending on odor release and sensing height.
We simulated Drosophila-inspired virtual agents performing source localisation in these plume recordings. The agents turned and moved depending on the odor signal they received in each timestep. Direction-sensing (DS+) agents turned towards the odor when odor-flow (lagged correlation between antennae) was above a threshold. Otherwise, they followed the DS- strategy: turn and move in upwind direction when odor concentration is above a threshold; or pursue a random walk below threshold.
DS+ agents reached the source in 77% of the trials in NB plumes, and 45% in FS conditions, while DS- achieved 74% success rate in NB and 30% in FS conditions, after optimization of all parameters for each combination.
Our results suggest that odor-flow detection via stereo antennae could be most beneficial in NB conditions, consistent with walking flies, and to a lesser amount in FS conditions. We plan to validate these strategies using robotic agents, with a view toward improving real-world odor source navigation.
References
[1] Kadakia et al. (2021) biorxiv https://doi.org/10.1101/2021.09.29.462473
[2] Haag et al. (2004) PNAS https://doi.org/10.1073/pnas.0407368101
[3] Connor et al. (2018) Exp Fluids https://doi.org/10.1007/s00348-018-2591-3
Funding
NSF Next Gen. Networks for Neuroscience #2014217 (JPC, ACT) & MRC #MR/T046759/1 (MS, SS)
EU H2020 HBP SGA3 #945539 (MS)