Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bayesian Decentralized Decision-making for Multi-Robot Systems: Sample-efficient Estimation of Event Rates

Gabriel Aguirre, Simay Atasoy Bingöl, Heiko Hamann, Jonas Kuckling

TL;DR

The paper tackles identifying the safer of two areas under unknown stochastic hazards in a multi-robot swarm. It introduces a decentralized Bayesian framework where each robot privately models interarrival times of hazardous events using a Weibull-exponential assumption with an inverse-gamma conjugate prior, and reaches consensus via a DMMD-based mechanism. Results show that DMMD-driven strategies can achieve correct and faster consensus with reduced exposure compared to a baseline, though belief sharing can induce faster but sometimes biased decisions. The work demonstrates a path toward risk-aware, sample-efficient exploration benchmarks for dynamic environments and suggests directions for handling censored data and extending to multiple options. Overall, it offers a principled approach to combining Bayesian inference with swarm-level majority dynamics for safe autonomous exploration.

Abstract

Effective collective decision-making in swarm robotics often requires balancing exploration, communication and individual uncertainty estimation, especially in hazardous environments where direct measurements are limited or costly. We propose a decentralized Bayesian framework that enables a swarm of simple robots to identify the safer of two areas, each characterized by an unknown rate of hazardous events governed by a Poisson process. Robots employ a conjugate prior to gradually predict the times between events and derive confidence estimates to adapt their behavior. Our simulation results show that the robot swarm consistently chooses the correct area while reducing exposure to hazardous events by being sample-efficient. Compared to baseline heuristics, our proposed approach shows better performance in terms of safety and speed of convergence. The proposed scenario has potential to extend the current set of benchmarks in collective decision-making and our method has applications in adaptive risk-aware sampling and exploration in hazardous, dynamic environments.

Bayesian Decentralized Decision-making for Multi-Robot Systems: Sample-efficient Estimation of Event Rates

TL;DR

The paper tackles identifying the safer of two areas under unknown stochastic hazards in a multi-robot swarm. It introduces a decentralized Bayesian framework where each robot privately models interarrival times of hazardous events using a Weibull-exponential assumption with an inverse-gamma conjugate prior, and reaches consensus via a DMMD-based mechanism. Results show that DMMD-driven strategies can achieve correct and faster consensus with reduced exposure compared to a baseline, though belief sharing can induce faster but sometimes biased decisions. The work demonstrates a path toward risk-aware, sample-efficient exploration benchmarks for dynamic environments and suggests directions for handling censored data and extending to multiple options. Overall, it offers a principled approach to combining Bayesian inference with swarm-level majority dynamics for safe autonomous exploration.

Abstract

Effective collective decision-making in swarm robotics often requires balancing exploration, communication and individual uncertainty estimation, especially in hazardous environments where direct measurements are limited or costly. We propose a decentralized Bayesian framework that enables a swarm of simple robots to identify the safer of two areas, each characterized by an unknown rate of hazardous events governed by a Poisson process. Robots employ a conjugate prior to gradually predict the times between events and derive confidence estimates to adapt their behavior. Our simulation results show that the robot swarm consistently chooses the correct area while reducing exposure to hazardous events by being sample-efficient. Compared to baseline heuristics, our proposed approach shows better performance in terms of safety and speed of convergence. The proposed scenario has potential to extend the current set of benchmarks in collective decision-making and our method has applications in adaptive risk-aware sampling and exploration in hazardous, dynamic environments.

Paper Structure

This paper contains 14 sections, 2 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Problem Definition
  4. Method
  5. Belief model
  6. Decision algorithms
  7. Robot controller
  8. Dissemination and observation times
  9. Experimental Setup
  10. Results
  11. Belief model convergence
  12. Accuracy and consensus time
  13. Observation efficiency
  14. Discussion and Conclusion

Figures (3)

  • Figure 1: Robot arena (zoomed in): red and blue event areas, central nest zone (green circle), and transition space (white).
  • Figure 2: Histograms of measured interarrival times (a) in the blue event area and (b) in the red event area for the easy environment, accumulated by the whole swarm. Measurements were taken during a typical run with event rates $\lambda_B=1/(2\times 10^4)$ and $\lambda_R=1/10^4$. We additionally plot the Weibull distribution corresponding to these event rates ("True distribution") and the Weibull distribution fitted to the measurements. (c) Comparison of number of observations over the whole swarm for: DMMD (standard), communication-free baseline, and DMMD approach with a higher prior; lighter colored boxes give performance in easy environment, darker boxes give performance in difficult environment.
  • Figure 3: Decision accuracy of the swarm across all experiments. (a) and (b) show the decision accuracy at termination time for the two environments, whereas (c) shows the decision accuracy over time in the easy environment.