Adaptive sampling of lion accelerometer data

The ability to track and monitor animals in the wild presents a challenge in both hardware and software development. This is a challenge facing all sensor networks, where there is a requirement for more computation to take place at the nodes whilst expending minimal battery life. In this paper we look at applying reinforcement learning on accelerometer data collected from a lion tag to learn a sampling control policy that adapts according to a lion’s behaviour. The results show that it is possible to reduce battery consumption by 73 % with a reconstruction accuracy of 51 %.