Generic Component-Based Mission-Centric Energy Model for Micro-Scale Unmanned Aerial Vehicles
The trend towards the usage of battery-electric unmanned aerial vehicles needs new strategies in mission planning and in the design of the systems themselves. To create an optimal mission plan and take appropriate decisions during the mission, a reliable, accurate and adaptive energy model is of utmost importance. However, most existing approaches either use very generic models or ones that are especially tailored towards a specific UAV. We present a generic energy model that is based on decomposing a robotic system into multiple observable components. The generic model is applied to a swarm of quadcopters and evaluated in multiple flights with different manoeuvres. We additionally use the data from practical experiments to learn and generate a mission-agnostic energy model which can match the typical behaviour of our quadcopters such as hovering; movement in x, y and z directions; landing; communication; and illumination. The learned energy model concurs with the overall energy consumption with an accuracy over 95% compared to the training flights for the indoor use case. An extended model reduces the error to less than 1.4%. Consequently, the proposed model enables an estimation of the energy used in flight and on the ground, which can be easily incorporated in autonomous systems and enhance decision-making with reliable input. The used learning mechanism allows to deploy the approach with minimal effort to new platforms needing only some representative test missions, which was shown using additional outdoor validation flights with a different quadcopter of the same build and the originally trained models. This set-up increased the prediction error of our model to 4.46%.