Adam Cataldo Edward Lee Shankar Sastry Soft Walls

Adam Cataldo Edward Lee Shankar Sastry Soft Walls http: //softwalls. eecs. berkeley. edu Introduction Objections to Soft Walls is a technological response to the September 11, 2001 hijackings. The Soft Walls strategy is to store a 3 D database of “no-fly zones”, or restricted airspace, on-board each aircraft and enforce these no-fly zones using an avionics control system. Each aircraft will have its own Soft Walls system. Also, the database will require a digital signature to update the no -fly zones so that the system is non-hackable. • Reducing pilot authority is dangerous. Pilots need the authority to respond to emergencies, including unexpected weather conditions, possible collisions with other aircraft or other obstacles, turbulence, on-board equipment failures, fires, or other problems. A pilot’s responsibility, however, extends beyond the craft, crew and passengers to the people on the ground. No on-board emergency is severe enough to justify endangering large numbers of people on ground. Previous Results • The crew should have an override. Design Objectives The Soft Walls controller does not remove pilot input when the aircraft approaches a no-fly zone. Instead the controller adds a bias to the pilot input. A pilot who approaches a no-fly zone and holds steady will be turned away from the no-fly zone until it is safe to let the aircraft fly straight. A pilot who chooses to turn away faster can do so. A pilot who tries to fly into the no-fly zone will be unsuccessful. Through this, Soft Walls will maximize pilot authority subject to the constraint that no-fly zones are enforced. This will give the pilot more maneuverability in an emergency. Pilot holds steady Pilot turns away from no-fly zone Pilot tires to fly into no-fly zone No-fly zone Control Applied 25 October 2021 The surest way to make the Soft Walls system effective is to prohibit override in any form. Manual override on the aircraft is certainly out of the question. • GPS is vulnerable to attacks. GPS signals are vulnerable to jamming, where a malicious party transmits a radio signal that swamps the one of interest, making it impossible to receive reliably, and spoofing, where a malicious party transmits a radio signal that masquerades as the radio signal of interest, hoping that it will be picked up instead of the legitimate signal. GPS signals currently contain encrypted channels that make spoofing extremely difficult. Since GPS cannot be spoofed, jamming can be reliably detected. A video game simulation of Soft Walls using Ptolemy II For simple models of the aircraft dynamics, we have applied analytical methods to derive provably safe control laws, as well as numerical which approximate provably safe control laws. Hybrid System Challenges An “optimal” controller in Soft Walls is one which always pushes away from the no-fly zone. To blend the control in gradually, we would like to increase the control bias from zero to the optimal bias when as the aircraft approaches the nofly zone. This can be dangerous; a pilot who fights the control input while it is between zero and the maximum bias can move the aircraft to a position where the optimal bias suddenly changes direction. This can cause chatter along a switching boundary, and may destabilize the aircraft. right bias required bias right no bias required The Research Problem Soft Walls is an example of a collision avoidance problem, where a “collision” occurs if the aircraft enters the no-fly zone, and we try to design a controller which will prevent collisions. Safety is critical in this application, so we must be able to prove that no collision can occur. For any controller, there may be some states for which we cannot prevent a collision. We must also be able to identify these dangerous states. left bias required bias left In this particular case, we can add states to ensure that this switching never occurs, while at the same time guaranteeing that collision avoidance. The end result is a “smooth” control law which guarantees safety, without much additional complexity added to the controller. An open research question is how to generalize this to other collision avoidance systems. Another important hybrid systems concept in Soft Walls is that of abstraction. That is, can we abstract our system model into a form over which it is easier to design a safe control law?