Cooperative localization by dual foot-mounted inertial sensors and inter-agent ranging

John-Olof Nilsson, Dave Zachariah, Isaac Skog, Peter Händel

TL;DR

This paper presents a novel, partially decentralized system architecture and sensor fusion approach for cooperative localization using foot-mounted inertial sensors and inter-agent ranging. It significantly reduces computational and communication costs, making robust, infrastructure-free localization feasible for groups of agents.

What Problem Does This Paper Address?

Accurate, infrastructure-free localization is crucial for many applications, especially in environments where GPS is unavailable (e.g., indoors, urban canyons). While foot-mounted inertial sensors offer robust pedestrian dead reckoning, and inter-agent ranging provides cooperative information, combining them effectively presents significant challenges. These include managing the high dimensionality of the system state, dealing with non-collocated sensors (left and right foot), and handling the often non-Gaussian and spatially correlated errors of ranging measurements. Existing approaches often struggle with computational complexity, communication bandwidth, or robustness.

Methods

The authors propose a partially decentralized system architecture that divides the inertial navigation process into two stages:

  1. Step-wise Inertial Navigation: This is performed locally on each agent's foot-mounted units. It leverages Zero-Velocity Updates (ZUPTs) to correct for drift. Crucially, it introduces a "step-wise dead reckoning" approach where the displacement and heading changes over a step are used to update the agent's global position and heading. This process is designed to be computationally light and requires minimal data transfer.
  2. Step-wise Dead Reckoning: This stage uses the output from the inertial navigation to maintain a global estimate of the agent's position and heading. This is performed on a local processing device within the agent.

For sensor fusion, they introduce:

  • Marginalization: This technique is used to handle the high dimensionality of the global state (positions and headings of all agents' feet). It allows for efficient updates by focusing only on the relevant parts of the state space for each measurement.
  • Range Constraints: To fuse information from the two feet of an individual agent, they exploit the fact that their spatial separation is bounded. This is implemented using sampling-based updates and projections.
  • Sampling-Based Updates for Inter-Agent Ranging: To robustly handle the non-Gaussian and potentially heavy-tailed errors of ranging measurements, they employ a sampling-based approach. This allows for more flexible likelihood functions than traditional Kalman filters, improving robustness to outliers.

Datasets

Not discussed in the paper. The evaluation is based on simulations and a real-time system implementation using synthetic ranging data generated from a Ubisense system.

Key Insights and Ideas

  • Decentralization is Key: A fully centralized or decentralized approach is suboptimal. A hybrid approach, where inertial navigation is localized and dead reckoning is performed locally, significantly reduces communication and computation.
  • Step-wise Processing: Breaking down the inertial navigation into discrete steps (steps of gait) allows for efficient updates and manageable state estimation.
  • Marginalization for Scalability: By marginalizing out irrelevant states, the system can scale to a larger number of agents without prohibitive computational cost.
  • Range Constraints for Intra-Agent Fusion: The bounded separation of an agent's feet provides a valuable constraint that can be used for robust fusion without additional sensors.
  • Sampling for Robust Ranging: Using sampling-based updates for inter-agent ranging makes the system resilient to non-ideal measurement conditions (e.g., heavy tails, outliers).

Key Results

  • Communication Reduction: The proposed architecture reduces communication bandwidth by approximately two orders of magnitude compared to a naive centralized approach.
  • Computational Cost: The system is designed to be computationally manageable for up to platoon-sized groups of agents.
  • Simulation Results:
    • In a straight-line march scenario (worst-case), the absolute localization RMSE grows slower with the number of agents, and the error growth is reduced by a factor of $\sqrt{N}$ (where N is the number of agents).
  • Relative position errors are bounded by inter-agent ranging.
  • In a scenario with static agents (best-case), both absolute and relative errors are bounded.
  • Real-time Implementation: A functional system was demonstrated with multiple agents in a harsh industrial environment, achieving meter-level accuracy over operation times of tens of minutes.

Limitations and Future Work

  • Initialization and Time Synchronization: The paper notes that initialization and time synchronization are not considered in detail, referring to other works for solutions.
  • Larger Groups: For groups larger than a platoon, a cellular structure might be needed.
  • Anchor Nodes: While the system is infrastructure-free, the best-case scenario involves static agents acting as implicit anchor nodes, which improves performance.
  • Future Work: The paper suggests that the ranging updates can be extended to incorporate information from other sources, such as anchor nodes or pressure measurements. The integration of voice radio communication and 3D audio into the localization system has also been explored.

Summary updated

Abstract

The implementation challenges of cooperative localization by dual foot-mounted inertial sensors and inter-agent ranging are discussed and work on the subject is reviewed. System architecture and sensor fusion are identified as key challenges. A partially decentralized system architecture based on step-wise inertial navigation and step-wise dead reckoning is presented. This architecture is argued to reduce the computational cost and required communication bandwidth by around two orders of magnitude while only giving negligible information loss in comparison with a naive centralized implementation. This makes a joint global state estimation feasible for up to a platoon-sized group of agents. Furthermore, robust and low-cost sensor fusion for the considered setup, based on state space transformation and marginalization, is presented. The transformation and marginalization are used to give the necessary flexibility for presented sampling based updates for the inter-agent ranging and ranging free fusion of the two feet of an individual agent. Finally, characteristics of the suggested implementation are demonstrated with simulations and a real-time system implementation.
BibTeX
@article{nilsson2013cooperative,
  author = {John-Olof Nilsson and Dave Zachariah and Isaac Skog and Peter Händel},
  title = {{Cooperative localization by dual foot-mounted inertial sensors and inter-agent ranging}},
  journal = {EURASIP J. Adv. Signal Process.},
  year = {2013},
  url = {https://openreview.net/forum?id=JyuhwL8pWY},
}
Figure