Flow structure decides whether a robot searching for odour should follow the wind

The gist: A wind-derived heading can guide a robot towards an odour source even after the wind reverses, but its reliability depends on the plume’s transport history<ref:2610.20970#pg6>

Reversing Wind Defeats Wind-Following

Holding an earlier heading can protect a searcher from this error<ref:2610.20970#pg4> On a Unitree Go1 quadruped searching for ethanol between alternating fan banks, a held heading reached the source while the comparison controllers failed to settle there<ref:2610.20970#pg4> The comparison controllers were surge-cast, a published moth model [14], reliability-modulated surge [15], and infotaxis [16]<ref:2610.20970#pg4> The rig measurements define the physical conditions behind this result<ref:2610.20970#pg5> The comparison controllers initially approach the source and lose ground after a reversal<ref:2610.20970#pg5> WindReacquire continues closing range<ref:2610.20970#pg5>

Transport Geometry and Sign Conditions

An exact geometric condition links heading correctness to net odour transport<ref:2610.20970#pg4> The source lies back along that displacement, whereas a wind sensor reports only the local wind now<ref:2610.20970#pg4> A held heading reaches the source through real wind reversals from a favourable start<ref:2610.20970#pg4> Under uniform advection, any mixture of two wind directions separated by σ Measured sign errors must come from departures from that flow or from perfect sensing<ref:2610.20970#pg9> The sign-error rate is 4.38 percentage points (pp) where nominal reversal alone predicts none<ref:2610.20970#pg9>

Separating Sensing Error from Plume Geometry

An angular decomposition identifies the dominant departure from the emission-heading approximation<ref:2610.20970#pg9> The plume share of the channel budget, E[ψ 2]/(E[η 2] + E[ψ 2]), is 76 % when controllers are pooled and 88 % for the hold-only variant<ref:2610.20970#pg9> This identity makes the reference axes explicit<ref:2610.20970#pg9>

Re-acquiring the Heading

WindReacquire adds one rule to a held heading, and the rule needs intermittent odour contact<ref:2610.20970#pg9> Re-acquisition is worth +46.5 pp with the conflict gate on and +44.0 pp with it off Re-acquiring from odour beats two controls that, on contact loss, replace the heading with a random direction or a random sign by 27.5 to 34.2 pp across the four control–plume pairs

Conclusion

A useful wind cue and a usable replacement process are distinct requirements for search in reversing flows The practical implication is to evaluate heading reliability and opportunities for renewed contact separately A correct sign was necessary for success on the hold-only variant, which never discards its heading

--- Page 5 ---

When a robot can trust reversing wind to find an odour source

Abstract

A wind-derived heading can guide a robot towards an odour source even after the wind reverses, but its reliability depends on the plume’s transport history<ref:2610.20970#pg4> On a quadruped initially facing the source, holding the initial heading reached it in every reversing-wind trial, whereas comparison controllers never settled there<ref:2610.20970#pg4> An exact geometric condition links heading correctness to net odour transport, and simulations separate sensing error from departures from nominal flow<ref:2610.20970#pg4> Where the alternating wind directions are too close for reversal alone to produce wrong-side headings, the wrong-side rate rises from 1.0% under uniform transport to 4.7% in a stochastic flow<ref:2610.20970#pg4> Re-acquiring headings from new odour contact improves simulated search when contact is intermittent<ref:2610.20970#pg4> These results identify distinct requirements for wind-guided search: a heading supported by the transport geometry and sufficient intermittent contact to replace it<ref:2610.20970#pg4>

Introduction

A robot searching for a gas leak detects intermittent filaments of odour, thin strands separated by clean air [1–7]<ref:2610.20970#pg4> Many search strategies follow an insect-inspired rule: surge upwind during contact and cast crosswind after contact is lost [8–10]<ref:2610.20970#pg4> These strategies use the current wind as a cue to the source direction<ref:2610.20970#pg4> A reversal can invalidate that cue while previously released odour is still arriving<ref:2610.20970#pg4> Holding an earlier heading can protect a searcher from this error<ref:2610.20970#pg4> On a Unitree Go1 quadruped searching for ethanol between alternating fan banks, a held heading reached the source while the comparison controllers failed to settle there (Fig. 1)<ref:2610.20970#pg4> However, the robot started facing the source, so its initial upwind heading was favourable<ref:2610.20970#pg4> The answer starts with transport<ref:2610.20970#pg4> A packet detected now has accumulated displacement under the wind along its path<ref:2610.20970#pg4> The source lies back along that displacement, whereas a wind sensor reports only the local wind now<ref:2610.20970#pg4> Their agreement depends on the packet’s history and on how the plume departs from the nominal flow<ref:2610.20970#pg4> We call storing a world-frame heading from the sensed wind a commitment<ref:2610.20970#pg4> Its sign is correct when the heading lies within a right angle of the true source bearing<ref:2610.20970#pg4> Correct sign is weaker than accurate alignment or eventual arrival<ref:2610.20970#pg4>

Results

Reversing wind defeats wind-following on a legged robot

Holding the initial heading preserved source arrival through wind reversals on the Go1<ref:2610.20970#pg5> The steady condition checked that each controller could solve the task before reversal was introduced<ref:2610.20970#pg5> WindReacquire’s settled distance was 0.31 m on average and within 0.39 m in every run, while the four windfollowing controllers settled between 1.05 m and 1.90 m on average<ref:2610.20970#pg5> The robot’s start therefore roughly doubles the held heading’s success, while following the current wind fails from either start

Simulation reproduces the collapse and isolates it

In the fixed source geometry of the silkmoth virtual-reality paradigm [14], a straight-line policy can solve the task without interpreting the wind<ref:2610.20970#pg6> The reactive family, five controllers that steer by the currently sensed wind, collapses to 0–5 % success across the tested release conditions Infotaxis is an exception to this collapse: in simulation it does not collapse at reversal, and its standing depends on where its prior over the source position is placed Two plume models driven by the same nominal wind process isolate the role of plume transport

When a wind-derived heading faces the source

Net packet displacement gives an exact geometric test for heading correctness<ref:2610.20970#pg4> A simpler approximation places the source on the upwind axis at emission<ref:2610.20970#pg4> Small wind separations isolate errors that nominal reversal alone cannot explain Each simulated measurement runs on a seed block, a disjoint range of random seeds that is never reused between selection and testing The sign-error rate is 4.

1. Improved Heading Reliability via Stored Commitment: The AI system can maintain a heading even after wind reversals by implementing a rule that stores an upwind heading on odour contact, rejects conflicting wind readings, and replaces the heading after odour contact is lost. This allows the robot to reach it in every reversing-wind trial when compared to controllers that never settled there.

2. Improved Search Strategy via Intermittent Contact: The system can improve simulated search when contact is intermittent by employing a mechanism where it learns that re-acquiring headings from new odour contact improves simulated search. This suggests the AI should be trained to prioritize re-acquisition of a heading upon re-detection.

3. Enhanced Source Localization via Transport Geometry: The AI can utilize an exact geometric condition linking heading correctness to net odour transport, which is defined by the relationship where the source lies back along that displacement, enabling it to verify if a sensed heading is correct based on accumulated packet history.

4. Robustness Against Flow Field Changes: The system can be designed to recognize when a prediction fails across different flow regimes, as the paper shows that the same construction fails on nominal advection, even with that model’s own no-reversal kernel. This suggests developing a mechanism to detect and adapt to changes in the underlying flow field.

5. Adaptive Contact Replacement: The system can implement a dynamic replacement strategy where it learns that Re-acquiring from odour beats two controls that, on contact loss, replace the heading with a random direction or a random sign, allowing for cue-informed replacement based on intermittent odour contact.