Turbulence and Adaptive Control Policies
The best feature combination for staying in a rising column was vertical wind acceleration together with torques.
The Control Problem Inside Rising Air
Finding rising air is only part of autonomous soaring. After a glider finds a rising column, its controller must help the glider remain inside that changing flow long enough to benefit from it. The central result here is that vertical wind acceleration combined with torques provided the best feature combination for staying in a rising column.
The controller is not merely asking, “Is there a thermal?” It is also asking, “What bank-angle choice will help me remain within the rising flow?”
Reading the Flow Through Two Signals
The strongest feature set contained only two kinds of information: vertical wind acceleration and torques. Together, these features provide information about the gradient of vertical wind velocity in two different directions. That information can support a choice between changing the bank angle to turn and leaving the bank angle unchanged to continue along the same course.
Choosing between a turn and a continued course
A controller receives vertical wind acceleration and torque information while the glider is in rising air. How can these signals support a control decision?
Combine the observations: The controller uses vertical wind acceleration together with torques rather than relying on only one kind of signal.
Interpret directional information: Together, the features provide information about the gradient of vertical wind velocity in two different directions.
Select a bank response: The controller can use that information to choose between changing the bank angle to turn and leaving the bank angle unchanged to continue along the same course.
Support position in the flow: The purpose of the decision is to help the glider remain within the rising column.
The feature combination is useful because it connects sensed changes in the flow to a bank-angle decision that can help maintain the glider's position within rising air.
Policies Under Weak and Strong Turbulence
A policy that works in one turbulence regime cannot simply be assumed to work equally well in every other regime. In weak turbulence, fluctuations give the controller more time to react. In strong turbulence, wind and glider velocities change rapidly, leaving less time to react. The amount of control possible is therefore lower when fluctuations are stronger, and strong turbulence leads to more conservative policies.
The policies are not completely unrelated across turbulence levels. Some responses are shared. When the controller senses negative wind acceleration, the common response is to bank sharply toward the wing with the higher lift. When it senses large positive wind acceleration with no torque, the common response is to do nothing. The policies differ in how aggressively they bank in other situations.
| Observed situation | Common response described in the results | What varies across turbulence levels |
|---|---|---|
| Negative wind acceleration | Bank sharply toward the wing with the higher lift | The broader aggressiveness of the policy |
| Large positive wind acceleration with no torque | Do nothing | Other situations may produce different banking intensity |
| Strong turbulence generally | Use a more conservative policy | Less control is possible because changes happen rapidly |
Shared responses and turbulence-dependent differences
Thresholds as Policy-Switch Signals
A controller can use a chosen threshold in vertical wind acceleration as a boundary between control regimes. When the measured acceleration crosses that threshold, the event signals that the current style of control may no longer be appropriate. The controller can then adjust its policy instead of treating every turbulence condition as identical.
Interpreting a threshold event
A controller is using one policy style and detects that vertical wind acceleration has crossed its chosen threshold. What should the event mean to the controller?
Treat the crossing as information: The crossing is not merely another measurement; it is a signal that the current turbulence or flow condition may differ from the condition for which the current control style is appropriate.
Reconsider the policy: The controller can use the event to decide that its policy should be adjusted rather than continuing to treat all conditions identically.
Change the response style: The adjusted policy can account for the changed control conditions, including the possibility that stronger turbulence leaves less time to react.
Threshold crossing provides a practical trigger for adaptive control: observe the event, reconsider the current policy, and switch to a more suitable response style.
Inside a Thermal or Between Thermals
The useful information depends on the control problem. For remaining inside a thermal, vertical wind acceleration and torques were especially useful because they support bank-angle decisions within a changing rising flow. Traveling between thermals is a different problem. For covering large distances, such as in cross-country gliding and bird migration, angle of attack was useful.
| Control problem | Especially useful information | Main control purpose |
|---|---|---|
| Remaining inside a thermal | Vertical wind acceleration and torques | Support bank-angle choices that help keep the glider within the rising flow |
| Traveling between thermals | Angle of attack | Support covering large distances, including cross-country gliding and bird migration |
The Adaptive Feedback Path
A useful way to organize the controller is as a feedback path. It senses vertical wind acceleration and torques, interprets those signals as information about the changing flow, selects a policy appropriate to the turbulence regime, and makes a bank-angle choice. If vertical wind acceleration crosses a chosen threshold, that event can trigger a policy adjustment. This keeps sensing, policy selection, and action connected to the changing environment.
Mistakes in Interpreting the Policies
Treating thermal detection as the complete soaring task.
The controller must also remain inside the rising column long enough to benefit from it.
Fix:
Evaluate whether the sensed features support useful bank-angle decisions for maintaining position within the changing flow.Using one fixed policy for every turbulence regime.
Strong turbulence changes wind and glider velocities rapidly, leaving less time to react and reducing the amount of control possible.
Fix:
Allow the policy to become more conservative in strong turbulence and consider threshold crossings as signals for policy adjustment.Assuming that all useful soaring features serve the same purpose.
Remaining inside a thermal and traveling between thermals are different control problems.
Fix:
Associate vertical wind acceleration and torques with staying within rising flow, and angle of attack with covering large distances between thermals.Interpreting every measured situation as requiring an aggressive turn.
The common response described for that situation is to do nothing.
Fix:
Distinguish shared responses from situations where the policy's banking aggressiveness differs.
Apply the Policy Logic
Explain how an adaptive glider controller should respond conceptually when it detects that vertical wind acceleration has crossed a chosen threshold while turbulence is becoming stronger.
Hints
- Start by identifying what the threshold crossing tells the controller about the current control regime.
- Explain why strong turbulence changes the amount of control available.
- State how the controller's policy should change, rather than naming a precise bank angle that is not provided here.
Separating two soaring decisions
A glider has found rising air and must decide whether to focus on staying within it or on traveling a large distance toward another thermal. Which information should be emphasized for each task?
Identify the task: Staying within a thermal requires maintaining position in a changing rising flow, whereas traveling between thermals requires covering a large distance.
Choose the inside-thermal features: For remaining inside the thermal, emphasize vertical wind acceleration and torques because they were the strongest feature combination for supporting bank-angle choices.
Choose the between-thermal feature: For traveling between thermals over large distances, angle of attack was useful.
The correct feature emphasis depends on the phase of soaring: vertical wind acceleration and torques for remaining inside a thermal, and angle of attack for long-distance travel between thermals.
Key Takeaways
- Vertical wind acceleration together with torques was the best feature combination for staying in a rising column.
- These signals provide information about the gradient of vertical wind velocity in two directions and support bank-angle choices.
- Strong turbulence gives the controller less time to react, so more conservative policies are appropriate than under weak turbulence.
- A threshold crossing in vertical wind acceleration can signal that the current control style should be adjusted.
- Remaining inside a thermal and traveling between thermals are different problems: the former emphasizes vertical wind acceleration and torques, while the latter can use angle of attack for large-distance travel.
Key Takeaways
- The best feature combination for staying inside rising air was vertical wind acceleration plus torques.
- The combination helps connect flow information to bank-angle decisions, such as turning or continuing on the same course.
- Weak and strong turbulence require different policy styles because strong turbulence leaves less time to react.
- Threshold crossings can provide events that trigger adaptive policy changes.
- The information useful for remaining inside a thermal is not identical to the information useful for traveling between thermals.