Laser feedback based self-mixing interferometry (SMI) has been demonstrated for diverse metric sensing applications. Typically, SMI sensors are based on such laser diodes (LDs) which provide mono-modal emission resulting in SMI signals in which each interferometric fringe occurs due to change in optical path length of λ/2, where λ is emission wavelength. However, in case multiple laser modes undergo SMI, then each mode contributes its own set of fringes. As LDs can emit multiple modes under variable operating conditions, so, non-detection of multiple SMI modes can cause drastic increase in measurement error due to wrong interpretation of fringes. Previously, detection of multiple laser modes undergoing SMI was achieved by adding spectroscopic instruments to the SMI setup. This, however, compromises the inherent simplicity of SMI sensing. In this work, an automatic SMI based multi-modality detection method is proposed which is able to detect if multiple modes of deployed LD are undergoing SMI and are contributing additional fringes within the SMI signal under variable sensing conditions. Such detection enables correct interpretation of SMI fringe count and can be used to signal occurrence of modehopping or secondary mode excitation. The method uses an artificial neural network, able to automatically identify uni-, bi-, or tri-modal SMI signals. Two different LDs (emitting at 637 nm and 650 nm) were used to acquire 131 experimental uni-, bi-, and tri-modal SMI signals for variable operating conditions and target vibration amplitude. The proposed system has achieved modality detection accuracy of 98.57% on 70 unseen experimental SMI signals.

No abstract

Continual learning is essential for all real-world applications, as frozen pre-trained models cannot effectively deal with non-stationary data distributions. The purpose of this study is to review the state-of-the-art methods that allow continuous learning of computational models over time. We primarily focus on the learning algorithms that perform continuous learning in an online fashion from considerably large (or infinite) sequential data and require substantially low computational and memory resources. We critically analyze the key challenges associated with continual learning for autonomous real-world systems and compare current methods in terms of computations, memory, and network/model complexity. We also briefly describe the implementations of continuous learning algorithms under three main autonomous systems, i.e., self-driving vehicles, unmanned aerial vehicles, and robotics. The learning methods of these autonomous systems and their strengths and limitations are extensively explored in this article.