Control Theory Lachlan Blackhall and Tyler Summers Control Theory • Advanced control methods are model based Controller u System Dynamics y • Use a mathematical model of the system to design controllers State Space Models • Inputs, outputs describe external behavior of system • State variables describe internal behavior of system • Mathematical model: dx x X Rn f (x,u) dt y h(x,u) u U R m y Y R p Optimal Control • Fundamental engineering problem: design the “best” controller given some constraints • Choose a function K :Y U to solve minimize l(x(t),u(t))dt 0 xÝ f (x,u) subject to y h(x,u) x X Rn u U R m y Y R p Optimal Control • Optimal control problems are hard – Infinite-dimensional, non-convex in general • Linear quadratic problems are solvable minimize (x T (t)Qx x(t) u (t)Qu u(t))dt T 0 subject to xÝ Ax Bu y Cx x Rn u Rm y Rp Linear Quadratic Regulator • Assume y = x (i.e. C = I) • LQR = linear quadratic regulator • Solution: optimal cost given by 1 A P PA Qx PBQu B P 0 T (x T T (t)Qx x(t) u (t)Qu u(t))dt x (0)Px(0) T T 0 • Optimal controller linear in state 1 u(t) Qu BT Px(t) The Kalman Filter • Often not possible to measure x directly y Cx v where v is measurement noise • Estimate x from measurements y (choose a function F :Y X ) • Solution similar to LQR – State estimate xˆ is linear in measurements y Linear Quadratic Gaussian • LQG = LQR + Kalman Filter minimize T T (x (t)Q x(t) u (t)Qu u(t))dt x 0 subject to xÝ Ax Bu w y Cx v x Rn u Rm y Rp • Optimal solution: u(t) Qu1BT Pxˆ (t) ˆ from Kalman filter x LQR Example • Vectored thrust aircraft LQR Example • Equations of motion (Newton’s Laws) • Nonlinear! – We can linearize any nonlinear system about an equilibrium point LQR Example • Equilibrium point: • State space model Ý) 0 ( , xÝ, yÝ, u1 F1 u2 F2 m g LQR Example • Linear model AT P PA Qx PBQu1BT P 0 u(t) Qu1BT Px(t) LQR Example • Simulation Automotive • Many subsystems in modern cars use control principles. – http://www.youtube.com/watch?v=MfOgwr hJG8A - Volvo Collision Avoidance – http://www.youtube.com/watch?v=16Izr52l pFw&feature=related - Lexus auto park Automotive (cont.) • DARPA Challenge – Two challenges. • The first to drive unaided across the desert. • The second to drive unaided around a city while performing a number of common tasks like parking. – http://www.youtube.com/watch?v=BSS0MZvoltw • Google Self Driving Car – http://www.youtube.com/watch?v=64w-v-RJpk8 Aeronautical • Aircraft have been an obvious candidate for control systems given the complexity of these systems. • Autopilots are a obvious example. • Preventing the Dutch roll mode when landing was solved using a control system called a yaw damper. – http://www.youtube.com/watch?v=jtBYlwp6 ygU Aeronautical (cont.) Traditionally, the performance (manoeuvrability, etc…) and handling of an aircraft were limited by the stability properties of an aircraft. Modern control systems have solved this fundamental problem ensuring stability but allowing high performance. Modern fighter aircraft are actually unstable. A human pilot can no longer control the plane but a control system can make the system stable and high performance. Aeronautical (cont.) • Modern aircraft now have fly-by-wire control systems that include: – Autopilot – Yaw dampers – Vibration damping – Auto-landing – Flutter prevention Aeronautical (cont.) • Other aeronautical systems where control is used include – http://www.youtube.com/watch?v=96WePg cg37I – nano hummingbird – UAV collision avoidance – Space launch vehicles – Satellites