In this paper, we propose an algorithm for simultaneous indoor self-localization and Sound Source Localization (SSL) using a swarm of microphone-embedded-micro-quadrocopters (size 10cm). Micro-quadrocopters are extremely noisy, have low CPU power and cannot lift heavy equipment: the small payload of each micro-quadrocopter ( g) only allows us to equip it with one microphone in addition to the inbuilt motion sensors. To perform robust SSL despite these issues, we propose three functions: (1) Self-localization of the quadrocopters using sound landmarks placed in the environment, and simultaneous localization of unknown sound sources; (2) Sound source detection; (3) Distributed data fusion based on noisy information from all members of the swarm. To achieve these, we propose three algorithms that are robust to noise, can perform with a varying number of quadrocopters, and do not rely on GPS nor motion capture to allow indoor operations: (1) A sound-based Unscented Kalman Filter (UKF) for self-localization of each quadrocopter; (2) A peak-based algorithm for sound source detection; (3) A distributed SSL algorithm for swarms with consensus-based integration using a new filter termed Unscented Kalman Consensus Filter (UKCF). We evaluated the proposed methods in real world and in simulated environments. The preliminary results show that the sound-based UKF represents an improvement of 37 on position estimation precision compared to basic dead reckoning approaches, even when the theoretical assumptions are violated; the distributed UKCF gives an improvement of 85 on SSL compared to a single-sensor approach in simulation.
展开▼
