Overview
Small unmanned aircraft, such as gliders and fixed-wing drones, are limited by battery capacity and flight time. In nature, birds and glider pilots extend their range by exploiting atmospheric energy sources such as thermal updrafts and wind gradients. For autonomous aircraft, detecting and using these energy sources requires real-time decision-making under uncertainty.
Led by Aamir Ahmad at the University of Stuttgart’s Institute of Flight Mechanics and Flight Control, the project “Decision Making for Environmental Energy Exploitation with Small Aircraft” (2021–2025) investigated how small aircraft can autonomously detect and exploit atmospheric energy to extend endurance.
Key results
- Developed end-to-end deep reinforcement-learning and deep-learning methods for autonomous thermal soaring, enabling aircraft to learn policies that integrate updraft detection, localization, and exploitation.
- Implemented temporal convolutional networks and recurrent neural networks to estimate thermal updraft properties from sequences of local sensor measurements, allowing the aircraft to infer the presence and structure of thermals without explicit external maps.
- Demonstrated sim-to-real transfer: policies trained in simulation were successfully deployed on real fixed-wing soaring aircraft, achieving autonomous thermal detection and sustained altitude gain in outdoor flight tests.
- Advanced decision-making strategies for explore–exploit trade-offs in soaring, including when to search for new lift sources versus when to remain in a detected thermal.
- Integrated energy-aware flight planning and control, combining reinforcement learning with classical control and model-predictive approaches for robust, energy-efficient flight.
Why it matters
The project addresses a key limitation of small autonomous aircraft: limited flight time due to battery constraints. By enabling aircraft to harvest energy from the environment, it can significantly extend mission duration for applications such as environmental monitoring, wildlife observation, and surveillance. The methods also contribute to broader questions in decision-making under uncertainty, showing how learning-based approaches can be combined with classical control to produce robust, energy-aware autonomy.
This project produced the following peer-reviewed publications:
- Gall, C., Fichter, W., & Ahmad, A. (2024). End-to-End Thermal Updraft Detection and Estimation for Autonomous Soaring Using Temporal Convolutional Networks. In 2024 IEEE International Conference on Robotics and Automation (ICRA), 17875–17881. doi:10.1109/ICRA57147.2024.10611234
- Notter, S., Fichter, W., & Ahmad, A. (2023). Deep Reinforcement Learning Approach for Integrated Updraft Mapping and Exploitation. Journal of Guidance, Control, and Dynamics, 46(9), 1725–1739. doi:10.2514/1.G007234
- Gall, C., Fichter, W., & Ahmad, A. (2025). A Multi-Simulation Approach with Model Predictive Control for Anafi Drones. In IEEE European Conference on Mobile Robots (ECMR 2025). (Accepted; preprint available on arXiv.)
- Additional related work on autonomous soaring, thermal updraft estimation, and bioinspired energy harvesting by Aamir Ahmad and collaborators, published in venues such as Journal of Guidance, Control, and Dynamics, Journal of the Royal Society Interface, and conference proceedings (ICRA, ECMR).
