Deep Reinforcement Learning Approach for Integrated Updraft Mapping and Exploitation

SOARING aircraft can cover large distances without consuming fossil fuel or electric power by harvesting atmospheric energy. While large transport aircraft have long been taking advantage of upper winds to reduce their energy demand, small aerial vehicles capable of wing-borne flight can harvest energy by exploiting thermal updrafts in the lower atmosphere. The majority of these aerial vehicles use all-electric propulsion systems nowadays. However, the range and endurance of electrically propelled aircraft are limited. Consequently, a significant amount of work has gone into developing guidance and control strategies to exploit thermal updrafts in the last decades. To autonomously exploit a thermal, one has to locate an updraft first.We classify previously published approaches for the problem of mapping thermal updrafts into model-free methods and methods that employ a thermal updraft observation model.