Sensing Through Faults: Collective Perception by Imperfect Robot Swarms
Collective perception is a foundational problem in swarm robotics, in which the swarm must reach a consensus on a coherent representation of the environment. Beyond localization, a swarm's awareness of its surroundings helps detect ecological phenomena, assess structural surface conditions, or estimate pollution sources. However, past studies on the collective perception problem involve robots with perfect sensing or small numbers of faulty robots. Instead, this dissertation considers the problem of using swarm robots whose sensors are imperfect. It presents a probabilistic algorithm that helps the robots collectively decide the frequency of an environmental feature. The algorithm, derived from optimal estimation techniques and a decentralized Kalman filter, enables the swarm robots to make accurate frequency estimates even with severe sensor degradation. Further, this dissertation introduces an adaptive self-calibration algorithm for the robots to self-calibrate while estimating the environmental features. With self-calibration, robots unaware of the degradation severity outperform those without, even reaching performance levels of robots aware a priori of their sensor degradation. Finally, the presentation concludes with an extension of the work thus far by studying the problem of the sensor degradation level deteriorating over time. Specifically, the dissertation proposes a Bayesian algorithm that enables simultaneous sensor degradation tracking with collective perception.
Advisor: Professor Carlo Pinciroli
Committee: Professor Nitin Sankit, Professor Kevin Leahy, Professor Heiko Hamann
Zoom link: https://wpi.zoom.us/j/93112710279