Optimization of Collective Bayesian Decision-Making in a Swarm of Miniaturized Vibration-Sensing Robots

TL;DR

The paper tackles a binary inspection problem where a swarm of miniaturized vibration-sensing robots must determine whether the majority of tiles on a 1 m × 1 m surface are vibrating (f>0.5) using a distributed Bayesian framework.The approach combines a Beta-distributed belief over the fill-ratio with three all-to-all information-sharing strategies, including a novel soft feedback mechanism, and optimizes algorithm parameters via a Webots-based simulator.A two-stage optimization framework (noise-resistant PSO followed by grid search) calibrates the simulation to real robot behavior, enabling robust performance across environmental difficulties and swarm sizes.Key findings show that soft feedback reduces decision time without sacrificing accuracy, and optimized parameters outperform empirically chosen ones, supporting scalable, real-time inspection with mobile sensor swarms.

Abstract

Inspection of infrastructure using static sensor nodes has become a well established approach in recent decades. In this work, we present an experimental setup to address a binary inspection task using mobile sensor nodes. The objective is to identify the predominant tile type in a 1mx1m tiled surface composed of vibrating and non-vibrating tiles. A swarm of miniaturized robots, equipped with onboard IMUs for sensing and IR sensors for collision avoidance, performs the inspection. The decision-making approach leverages a Bayesian algorithm, updating robots' belief using inference. The original algorithm uses one of two information sharing strategies. We introduce a novel information sharing strategy, aiming to accelerate the decision-making. To optimize the algorithm parameters, we develop a simulation framework calibrated to our real-world setup in the high-fidelity Webots robotic simulator. We evaluate the three information sharing strategies through simulations and real-world experiments. Moreover, we test the effectiveness of our optimization by placing swarms with optimized and non-optimized parameters in increasingly complex environments with varied spatial correlation and fill ratios. Results show that our proposed information sharing strategy consistently outperforms previously established information-sharing strategies in decision time. Additionally, optimized parameters yield robust performance across different environments. Conversely, non-optimized parameters perform well in simpler scenarios but show reduced accuracy in complex settings.

Figures (15)

• Figure 1: The experimental setup with a fill ratio of $f = \frac{12}{25} = 0.48$. (a) Schematic overview of the setup is shown. The central PC uses the radio and camera data for analysis. Vibration-motors are attached on the bottom side of white tiles. (b) A snapshot from the overhead camera with detailed view (black square). The red dot markings indicate vibrating tiles. Each robot carries a unique AruCo marker for tracking. AruCo markers in the corners of the environment mark the boundaries.
• Figure 2: Finite state machine of robot. The robot alternates between the states random walk (exploration), random turn (collision avoidance), and observation (sensing).
• Figure 3: We use a revised and extended version of the Rovable robot. (a) The extended robot with IR sensor board. (b) Exploded 3D CAD view of the extended robot. (c) Electronic block diagram of the extended robot. The microcontroller (Atmel SAMD21G18) interfaces with the IMU (MPU6050), 2.4 GHz radio (nRF24L01+), motor-controllers (DRV8835), and ToF IR sensors (VL53L1X).
• Figure 4: We use a simpler robot model in simulation, with IR sensors directly simulated on the main PCB. Our simulated collision avoidance closely matches reality.
• Figure 5: Our optimization and evaluation process: Stage 1 optimizes joint algorithm parameters across three feedback strategies using noise-resistant Particle Swarm Optimization (PSO), utilizing the no-feedback strategy ($u^-$). Stage 2 employs a grid search to optimize soft feedback parameters $\kappa$ and $\eta$. Afterwards, all three feedback strategies are assessed in simulation and experiments.