Aerospace and Mechanical Engineering
Adjust Font Size: Normal Large X-Large

AEM faculty spotlight:

Bérénice Mettler

After driving the same route home time and again, we tend to go on autopilot of sorts. Not to be confused with Highway Hypnosis, where an individual loses driving ability, this form of autopilot allows us to keep making good decisions, though the driver may be unaware of the exact details of the drive home, after the fact. And if the situation suddenly demands more attention, the human on autopilot can snap into conscious awareness. Bérénice Mettler, an assistant professor with AEM, hopes to bring insights about these phenomena to the world of autonomous vehicle, through her study of the pilot-vehicle interaction. Below is a conversation with Professor Mettler, in which she discusses autonomous control for unmanned aerial vehicles and applying the adaptability of the human mind to flight control.

Bérénice Mettler

What types of research interest you?
The general research area I'm interested in is autonomy, mostly for agile aerial vehicles like helicopters are miniature airplanes. Autonomous aerial vehicle control is one of the area where robotic technologies will have the biggest impact in the near future. However, there are significant challenges that need to be solved, in order to enable some of these applications and fully benefit from autonomous control technologies. In short, the main challenge is understanding how to replace the human pilots - that we've used for nearly 100 years - with computers and sensors.

How do we replace the human pilot in flight?
There are some fundamental problems that we need to solve. The main one has to do with the idea of representing flight tasks so that a computer can solve them. We've been quite successful in solving autonomous flight tasks in areas of flight where the operation is sufficiently organized and structured. When a pilot prepares an airplane for landing, he or she follows a precise nearly mechanical procedure. Understanding this procedure can be help automate the process. If the computer were to fly the aircraft, there are a finite number of things that could happen where the computer would respond appropriately. However not every aspect of flight operation is as mechanical. If you think about humans and what we do in our daily lives, there are lots of mundane situations - how we control a car or bicycle, for example - where the process does not have an obvious structure. Life calls on us to be creative in how we solve many problems in the moment. Even if we're not really conscious of how we're doing it, these situations are not trivial to implement on a computer. They are problems that people have been studying under the heading of "artificial intelligence" or "AI," where the goal is to give computers intelligence, so they can handle more than a finite set of scenarios.

How do you give flight computers intelligence?
In order for the flight computer to determine how to guide the aircraft the flight tasks has to be converted to a mathematical problem the computer can solve. Some tasks are quite simple to represent mathematically - like landing or traveling between two airports - this is so because the rules and constraints governing airspaces act like an aerial roadway system. The flight task can be described as finding a path in that roadway system. Some tasks cannot be formulated in such a structured manner. To make the situation harder the information needed to solve the problem is often not directly available. Think of a reconnaissance operation in an urban environment to locate an object or person. This is now much less trivial since the problem does not have an obvious structure. In addition to generating a plan, the computer has to guide the aircraft to gather information to complete the mission (detect obstacles and determine a terrain map using imaging or radar-like sensors).

How does actually flying the aircraft factor in to this problem?
Even if the environment were known perfectly ahead of time, generating trajectories for an agile vehicle like a helicopter cannot be done without significant computational effort. This is a major issue, since the speed at which airplanes travel and the limited payload does not enable large computations to be performed in real time. With aerial vehicles a mistake can be catastrophic. Even if the problems can be formulated mathematically these formulations often do not lead to tractable algorithms - leading to my other direction of research.

Would you talk a bit about this area?
Medevac or combat pilots are trained to operate in challenging geographical environments and conditions under severe time constraints. These skills encompass a range of functions working in concert. Researchers have worked a great deal on human piloting skills, but mainly focusing at a lower level, like how well they can follow a pre-specified trajectory, rather than on the overall system responsible for autonomy. One reason why understanding human pilots is relevant is that some of the unknowns about human skills are analogous to the engineering challenges we faces with UAV autonomy. Similarly to computers, humans have limitations in how much “computations” they can perform online and how much information they can account for in these computations. These issues are central in understanding how pilots generate the flight plan in a first place and how they adapt the plan in real time. In recent years, with advances in brain-sensing tools, we have opportunities to better understand how the human brain performs these functions. Maybe it is then possible to learn from these studies how to solve these different challenges. In my lab we are studying autonomous flight both from a mathematical and from a human perspective.

What needs to happen to implement all these problems associated with autonomous flight?
There are big scientific and technological challenges that need to be solved in order to go from operating an aircraft autonomously in a structured task, to tasks in more complex environments and conditions. These are the tasks - search and rescue operations for instance - that will offer the highest pay off. They challenge the methodologies that we've used in the past, so we'll really have to go through some form of revolution, in terms of developing fundamentally new tools. In particular focusing on the system-wide understanding of control, perceptual and cognitive functions involved in agile operation. What makes this area fascinating is the duality between man and machines. We cannot improve technology much further without understanding humans. The new insights needed to enable autonomous flight will also help make human flight operation safer.

Last Modified: Tuesday, 03-Jan-2012 14:54:56 CST -- this is in International Standard Date and Time Notation