Exoskeletons and orthoses are wearable mobile systems providing mechanical benefits to the users. Despite significant improvements in the last decades, the technology is not fully mature to be adopted for strenuous and non-programmed tasks. To accommodate this insufficiency, different aspects of this technology need to be analysed and improved. Numerous studies have been trying to address some aspects of exoskeletons, e.g. mechanism design, intent prediction, and control scheme. However, most works have focused on a specific element of design or application without providing a comprehensive review framework. This study aims to analyse and survey the contributing aspects to the improvement and broad adoption of this technology. To address this, after introducing assistive devices and exoskeletons, the main design criteria will be investigated from a physical Human-Robot Interface (HRI) perspective. The study will be further developed by outlining several examples of known assistive devices in different categories. In order to establish an intelligent HRI strategy and enabling intuitive control for users, cognitive HRI will be investigated. Various approaches to this strategy will be reviewed, and a model for intent prediction will be proposed. This model is utilised to predict the gate phase from a single Electromyography (EMG) channel input. The outcomes of modelling show the potential use of single-channel input in low-power assistive devices. Furthermore, the proposed model can provide redundancy in devices with a complex control strategy.