A Plume Tracking Algorithm Based on Crosswind Formations

Thomas Lochmatter, Ebru Aydin Göl, Iñaki Navarro, Alcherio MartinoliDars 2010 · DOI

TL;DR

This paper introduces a novel multi-robot algorithm for tracking odor plumes by having robots form a "crosswind formation." This approach allows robots to collaboratively navigate upwind towards an odor source, outperforming previous methods by avoiding spatial competition and continuously sharing sensor data.

What Problem Does This Paper Address?

Locating odor sources is crucial for applications like search and rescue, safety operations, and demining. However, plumes are intermittent and difficult to track, especially in dynamic environments. Existing multi-robot approaches often suffer from robots competing for space, leading to performance degradation. This paper aims to solve this by developing a collaborative formation strategy that leverages multiple robots to efficiently and reliably track odor plumes.

Methods

The core of the proposed algorithm is a "crosswind formation" where robots maintain a line formation perpendicular to the wind direction. This formation allows them to stay centered within the plume while moving upwind. Key aspects of the method include:

  • Collaborative Sensing: Robots share their odor concentration and wind direction measurements with each other.
  • Relative Pose Awareness: Robots know the relative positions of their teammates.
  • Formation Control: A reactive control scheme uses virtual forces (upwind and crosswind) to keep the formation centered in the plume.
  • Upwind Force ($f_u$): Encourages robots to move upwind and maintain their relative downwind positions.
  • Crosswind Force ($f_c$): A weighted combination of attractive and repulsive forces based on odor concentration measurements from all robots. High concentrations pull the formation towards them, while repulsive forces prevent robots from getting too close.
  • Wind-Aligned Reference Frame: Calculations are performed in a reference frame defined by the wind direction, simplifying the control logic.
  • No Low-Level Collision Avoidance: The formation strategy inherently prevents robots from competing for space, eliminating the need for explicit low-level collision avoidance algorithms between robots.

Datasets

Not discussed in the paper. The experiments were conducted in a custom-built wind tunnel.

Key Insights and Ideas

  • Formation as a Solution to Competition: By designing a formation-based approach, the authors effectively eliminate the problem of robots competing for space, a significant limitation in previous multi-robot odor tracking methods.
  • Continuous Collaboration: The algorithm's strength lies in the continuous sharing of all sensor data and the use of this collective information for control. This allows for immediate feedback and correction.
  • Implicit Plume Reacquisition: The formation's ability to stay centered and continuously receive feedback means that a separate "plume reacquisition" phase, common in single-robot algorithms, is no longer necessary. Robots don't lose the plume.
  • Simplicity and Low Requirements: Despite its effectiveness, the algorithm is described as simple and having low requirements, particularly regarding relative localization.

Key Results

  • Close-to-Optimal Performance: Experiments with up to 5 real robots in a laminar flow wind tunnel demonstrated that the algorithm achieves close-to-optimal performance in terms of distance overhead and success rate.
  • 100% Success Rate: In experiments with a static source, the success rate was 100% across all tested starting positions and robot configurations (3 and 5 robots).
  • Extremely Low Distance Overheads: The distance overhead (traveled distance divided by shortest path to source) was consistently below 2%, and often below 1%, even including the initial formation phase.
  • Robustness to Moving Sources: The algorithm also performed well when tracking a source that moved in the crosswind direction, demonstrating its ability to adapt to dynamic scenarios.

Limitations and Future Work

  • Static and Laminar Flow Scenario: The current experiments were conducted in a controlled environment with laminar flow and no obstacles.
  • Moving Source Trajectories: While successful, calculating distance overhead for moving sources was not feasible with the defined metrics.
  • Future Work:
    • Testing in more complex scenarios (obstacles, turbulence, complex source motions, multiple sources).
  • Investigating different robot formations.
  • Reducing positioning requirements to direct-neighbor-only information.
  • Comparing with single-robot algorithms using multiple sensors.
  • Exploring the impact of failing robots.

Summary updated

BibTeX
@inproceedings{lochmatter2010plume,
  author = {Thomas Lochmatter and Ebru Aydin Göl and Iñaki Navarro and Alcherio Martinoli},
  title = {{A Plume Tracking Algorithm Based on Crosswind Formations}},
  booktitle = {Dars},
  year = {2010},
  doi = {10.1007/978-3-642-32723-0\_7},
}
Figure