Deep models optimization on embedded devices to improve the orientation estimation task at sea

The environmental monitoring task has greatly benefited from the improvements achieved in the robotics field. The enhancement of navigation and control algorithms, together with the use of performing, small and low-cost sensors, allows in fact to reduce the implementation costs while improving the system reliability. This is strongly supported by the developments of embedded hardware, smart computing devices able to collect and process data in real-time and in low-resource settings. Following the results obtained by DOES, this work aims at putting another step towards its deployment in live scenarios: we propose a study on the performances of DOES tested on embedded systems, using lighter backbone architectures and model optimization techniques.