Energy-Conserving Risk-Aware Data Collection Using Ensemble Navigation Network

The Data-collection Problem (DCP) models robotic agents collecting digital data in a risky environment under energy constraints. A good solution for DCP needs a balance between safety and energy use. We develop an Ensemble Navigation Network (ENN) that consists of a Convolutional Neural Network and several heuristics to learn the priorities. Experiments show ENN has superior performance than heuristic algorithms in all environmental settings. In particular, ENN has better performance in environments with higher risks and when robots have low energy capacity.