Learning to soar using atmospheric thermals
Soaring birds often rely on ascending thermal plumes (thermals) in the atmosphere as they search for prey or migrate across large distances. How soaring birds find and navigate thermals is unknown. This is a scenario where experiments are difficult to control and the strategies used by birds are difficult to infer. In this work, I used modern methods from artificial intelligence as tools to generate hypotheses for the strategies and mechanosensory cues that birds may use in order to soar effectively.
In Chapter 1, I describe how a technique from artificial intelligence, namely, reinforcement learning, is used to train virtual gliders with bird-like aerodynamic properties to navigate simulated convective turbulent flows. By experimenting with the learning environment, we find that gliders need to sense and respond to two cues in order to soaring effectively: the vertical wind acceleration and the velocity differences across the wings. The learning process also yields a strategy for soaring within thermals that relies on these two cues.
In Chapter 2, I describe the details of how lessons from the simulations were used to teach gliders to navigate atmospheric thermals in the field. Gliders of two-meter wingspan were equipped with a flight controller that precisely controlled the bank angle and pitch, modulating these at intervals with the aim of gaining as much lift as possible. A navigational strategy was determined solely from the gliders pooled experiences collected over several days in the field. The strategy relies on methods to accurately estimate the local vertical wind accelerations and the roll-wise torques on the glider, which serve as navigational cues. I show that vertical wind accelerations and roll-wise torques are effective mechanosensory cues for soaring and provide a navigational strategy applicable to autonomous soaring vehicles.