Micro aerial vehicles (MAVs) have become increasingly prominent in the last decade, with several sectors now routinely using this technology for applications such as filming, surveying and maintenance. A significant barrier towards further MAV technology adoption is the absence of reliable, lightweight autonomous navigation systems that can robustly operate in areas where global navigation satellite systems (GNSS) signals are not reliable. Flying insects are an order of magnitude smaller than MAVs and they can navigate between several sites of interest in large local neighbourhoods that span several kilometres. Fed by low resolution eyes and using neural processing circuits, the power con sumption of an insect's brain is several orders of magnitude lower than state-of-the-art robotic visual navigation systems. This formidable capability has inspired ethologists, neuroscientists and engineers to engage in a process of reverse engineering the key mechanisms involved in local insect navigation behaviours, with the ultimate goal of describing the complete underlying neural circuitry. In this thesis, recent advances in MAV technology are exploited as a means of evaluating candidate behavioural models that have only been deployed in simulation environments or on terrestrial robotic platforms. The hardware and software development of an aerial biorobot that is configured to test insect navigation models is described. This system features a quadcopter airframe, Pixhawk flight controller and selected interfacing ancillary avionics. The resultant platform has sufficient onboard processing power to flexibly deploy path integration and visual homing behavioural models. The biorobot also features an active mechanical view stabilisation system. The biorobot is first used to embody a recently proposed anatomically constrained path integration circuit. To this end, a biologically plausible matched filter visual odometry pipeline is implemented. The viewing direction, resolution and field of view of the visual input to this circuit is systematically altered and tested in a variety of natural scenes. This process enables the prescription of an optimal visual sensor configuration on the basis of empirical evidence. When the biorobot is deployed in a relatively flat environment with the optimal view configuration, a homing error drift rate of 1.5m per 100m is estimated. The biorobot subsequently supports an investigation into whether flying insects could use visual route following to overcome the drift issues associated with path integration. A robust procedure is developed and evaluated. It is found to be effective across distances of at least 30m, even in seemingly featureless environments such as empty arable fields. It is known that orientated bandpass filters exist in the early stages of the human vision system. Using a complex wavelet structural similarity algorithm, the orientated bandpass filter approach is adapted to a visual homing framework. This configuration is shown to double the catchment area and increase the discriminability of the snapshot model for view matching in natural scenes when it is compared to existing view matching techniques that operate in the spatial domain.