Autonomous Multi-Robot Search for a Hazardous Source in a Turbulent Environment

Branko Ristić, Daniel Angley, Bill Moran, Jennifer L. PalmerSensors 2017 · DOI

TL;DR

This paper presents an autonomous multi-robot search algorithm for locating hazardous sources in turbulent environments. It combines a novel Bayesian estimation technique with a formation control strategy to efficiently search large areas, outperforming previous methods by overcoming particle depletion issues and handling intermittent sensor readings.

What Problem Does This Paper Address?

Imagine a scenario where a toxic substance is accidentally released into the atmosphere. Quickly finding the source of this release is crucial for national security and public safety. This problem is particularly challenging because the substance's plume is often dispersed and broken up by turbulence, leading to sporadic and unreliable sensor readings. Traditional search algorithms struggle with these conditions, either requiring smooth concentration gradients or suffering from computational issues like particle depletion in more advanced methods. This paper tackles the problem of efficiently and autonomously searching for such a source using a team of robots, even when sensor data is intermittent and the environment is turbulent.

Methods

The core of the proposed solution lies in two main components:

  • Cognitive Search (Infotaxis) with Rao-Blackwell Dimension Reduction: Instead of relying on direct concentration gradients, the robots employ an "infotaxis" strategy, which guides them towards areas of highest information gain. This is underpinned by a Bayesian framework for estimating the source's parameters (location and release rate). The key innovation here is the use of a Rao-Blackwell dimension reduction method for parameter estimation. This technique analytically solves for the source's release rate, conditioned on its location, and uses a Monte Carlo method only for the location estimation. This approach significantly improves accuracy and avoids the particle depletion issues common in particle filter-based methods, especially for long-duration searches.
  • Scalable Formation Control: The robots operate in a coordinated, scalable formation. Their movement is controlled autonomously using an information-theoretic measure (entropy reduction) as the reward function, balanced with a travel cost. This allows the formation to adapt its size and shape, ensuring all-to-all communication while efficiently exploring the search area. The control vector includes linear and angular velocity, travel time, and formation scale increment.

Datasets

The paper evaluates its algorithm using two types of data:

  • Simulated Data: The algorithm's performance is extensively tested using simulated turbulent environments. The parameters for these simulations are detailed in Section 5.1.
  • Experimental Data: The algorithm is also validated using real-world experimental data collected from a large recirculating water channel. This dataset involves fluorescein dye released at a constant rate, providing a realistic representation of plume dispersion. The data consists of 340 frames of concentration field measurements.

Key Insights and Ideas

  • Information Gain Over Gradients: The infotaxis strategy, focusing on information gain rather than direct concentration gradients, proves effective in turbulent environments where plumes are fragmented.
  • Rao-Blackwell for Robustness: The Rao-Blackwell dimension reduction method for parameter estimation is a crucial advancement, overcoming the limitations of previous particle filter approaches by analytically handling part of the state space.
  • Scalable Formations for Communication: The ability of the robot formation to scale is vital for maintaining communication links between robots, especially in large search areas.
  • Intermittent Search Behavior: The observed robot displacement patterns, characterized by long non-sensing movements followed by shorter search phases, closely mimic natural foraging behaviors, suggesting an efficient strategy for this type of problem.

Key Results

  • 100% Success Rate: The algorithm demonstrated a 100% success rate in finding the hazardous source in both simulated and experimental scenarios.
  • Accurate Localization: The localization accuracy was determined by the termination criterion, with the RMS localization error found to be approximately √w, where w is the termination threshold.
  • Diminishing Returns with More Robots: Increasing the number of platforms beyond a certain point yielded diminishing returns in search time reduction.
  • Linear Search Time with Area: The mean search time grew approximately linearly with the size of the search area.
  • Travel Cost Impact: Increasing the travel cost led to shorter formation displacements, indicating a trade-off between exploration speed and energy expenditure.
  • Inverse Gaussian Search Time: The search times for both single and multi-robot scenarios were accurately modeled by an inverse Gaussian distribution.

Limitations and Future Work

  • Assumptions: The paper assumes an open search field without obstacles and relies on an all-to-all communication model.
  • Future Directions: The authors suggest several avenues for future research:
  • Distributed Estimation and Control: Exploring distributed, rather than centralized, approaches for parameter estimation and robot control.
  • Realistic Communication: Incorporating more realistic communication networks with limited bandwidth and time delays.
  • Search in Obstacle-Ridden Environments: Adapting the algorithm for environments with known or unknown maps and obstacles, which would require different dispersion models and potentially new localization and mapping capabilities.
  • Integrated Search, Localization, and Mapping: The ultimate goal is to enable robots to autonomously perform all three functions simultaneously.

Summary updated

Abstract

Finding the source of an accidental or deliberate release of a toxic substance into the atmosphere is of great importance for national security. The paper presents a search algorithm for turbulent environments which falls into the class of cognitive (<i>infotaxi</i>) algorithms. Bayesian estimation of the source parameter vector is carried out using the Rao-Blackwell dimension-reduction method, while the robots are controlled autonomously to move in a scalable formation. Estimation and control are carried out in a centralised replicated fusion architecture assuming all-to-all communication. The paper presents a comprehensive numerical analysis of the proposed algorithm, including the search-time and displacement statistics.
BibTeX
@inproceedings{ristic2017autonomous,
  author = {Branko Ristić and Daniel Angley and Bill Moran and Jennifer L. Palmer},
  title = {{Autonomous Multi-Robot Search for a Hazardous Source in a Turbulent Environment}},
  booktitle = {Sensors},
  year = {2017},
  doi = {10.3390/s17040918},
}
Figure