Robot Control Basics CS 685

Transcription

1 Robot Control Basics CS 685

2 Control basics Use some concepts from control theory to understand and learn how to control robots Control Theory general field studies control and understanding of behavior of dynamical systems (robots, epidemics, biological systems, stock markets etc.)

3 Control basics Basic ingredients - state of the system current position of the robot - dynamics behavior of the systems as a function of time (description how system state changes as a function of time) - system of differential equations x = v cosθ y = v sinθ θ = ω x = [x, y,θ] x = f( x, u ) - control input which can affect the behavior u = [v,ω] - controller which takes some function of the goal, the state

4 Control basics Basic ingredients - controller which takes some function of the goal, the state - y output, measurement of some aspect of the state Feedback control how to compute the control based on output (state) and the desired objective r + - u x y Difference equations (examples) x k+1 = f( x k, u k )

5 Simple control strategies Moving to a point go to a point Consider a problem of moving to a point (x,y) How to control angular and linear velocity of the mobile robot Linear velocity proportional to distance Angular velocity steer towards the goal Following a line steer toward a line Angular velocity proportional to the combination distance from the line and also to alignment with the line

6 Moving to a point Differential drive robot go from the current pose to desired point with coordinates [x,y ] T [x, y, ] T = tan 1 y x v = K v p (x x) 2 +(y y) 2 )! = K h ( ) [x,y ] T [x, y, ] T y x {I} {R} Source P. Corke: Robotics, Vision and Control. Springer

7 Moving to a line Equation of a line Shortest distance of the robot the line Orientation of the line [x, y, ] T {I} {R} ax + by + c =0 d = [a, b, c][x, y, 1]T p a2 + b 2 = tan 1 d = K d d K d > 0 Steer towards the line and align the robot with the line a b h = K h ( )! = K d d + K h ( )

8 Following a path Same as going to the point now sequence of waypoints x(t),y(t) [x,y ] T = tan 1 y y p x x e = K v (x x) 2 +(y y) 2 ) d [x, y, ] T {I} {R} d distance behind the pursuit point v = K v e + K i Z! = K h ( ) edt Source P. Corke: Robotics, Vision and Control. Springer

9 More suitable alternative Feedback control Use state feedback controller At each instance of time compute a control law Given the current error between current and desired position Δy

10 Kinematic Position Control The kinematic of a differential drive mobile robot described in the initial frame {x I, y I, θ} is given by, Δy I x! cosθ = y! sinθ! θ 0 0 v 0 ω 1 relating the linear velocities in the direction of the x I and y I of the initial frame. Let α denote the angle between the x R axis of the robots reference frame and the vector connecting the center of the axle of the wheels with the final position. R. Siegwart, I. Nourbakhsh

11 Motion Control (kinematic control) 3.6 The objective of a kinematic controller is to follow a trajectory described by its position and/or velocity profiles as function of time. Motion control is not straight forward because mobile robots are non-holonomic systems. However, it has been studied by various research groups and some adequate solutions for (kinematic) motion control of a mobile robot system are available. Most controllers are not considering the dynamics of the system R. Siegwart, I. Nourbakhsh

12 Motion Control: Feedback Control Set intermediate positions lying on the requested path. Given a goal how to compute the control commands for linear and angular velocities to reach the desired configuration Problem statement Given arbitrary position and orientation of the robot how to reach desired goal orientation and position [x, y,θ] [x g, y g,θ g ] Δy R. Siegwart, I. Nourbakhsh

13 Motion Control: Feedback Control, Problem Statement y R ω (t) start v(t) θ x R e goal Find a control matrix K, if exists k11 k12 k13 K = k with k ij =k(t,e) such that the control of v(t) and ω(t) k k v( t) = K e = K ω ( t) R x y θ drives the error e to zero. lim e( t) = 0 t R. Siegwart, I. Nourbakhsh

14 Motion Control: Kinematic Position Control The kinematic of a differential drive mobile robot described in the initial frame {x I, y I, θ} is given by, I x! y!! θ = cosθ sinθ 0 0 v 0 ω 1 where and are the linear velocities in the direction of the x I and y I of the initial frame. Let α denote the angle between the x R axis of the robots reference frame and the vector connecting the center of the axle of the wheels with the final position.

15 Kinematic Position Control: Coordinates Transformation Coordinates transformation into polar coordinates with its origin at goal position: Δy System description, in the new polar coordinates α For from for

16 Kinematic Position Control: Remarks The coordinates transformation is not defined at x = y = 0; as in such a point the determinant of the Jacobian matrix of the transformation is not defined, i.e. it is unbounded For the forward direction of the robot points toward the goal, for it is the backward direction. By properly defining the forward direction of the robot at its initial configuration, it is always possible to have at t = 0. However this does not mean that α remains in I 1 for all time t.

17 Kinematic Position Control: The Control Law It can be shown, that with the feedback controlled system will drive the robot to ρ, α, β = 0, 0, 0 The control signal v has always constant sign, the direction of movement is kept positive or negative during movement ( ) ( ) parking maneuver is performed always in the most natural way and without ever inverting its motion.

18 Question: How to select the constant parameters k s so as to achieve that the error will go to zero Digression eigenvectors and eigenvalues review

19 Eigenvalues and Eigenvectors Motivated by solution to differential equations For square matrices! " v w $! = A % " v w $ % Suppose solution have this form of exponentials For scalar ODE s Substitute back to the equation and denote x as then eq. above is

20 Eigenvalues and Eigenvectors eigenvalue eigenvector Solve the equation: x is in the null space of λ is chosen such that (1) has a null space Computation of eigenvalues and eigenvectors (for dim 2,3) 1. Compute determinant 2. Find roots (eigenvalues) of the polynomial such that determinant = 0 3. For each eigenvalue solve the equation (1) For larger matrices alternative ways of computation

21 For the previous example Eigenvalues and Eigenvectors We will get special solutions to ODE Their linear combination is also a solution (due to the linearity of ) In the context of diff. equations special meaning Any solution can be expressed as linear combination Individual solutions correspond to modes

22 Eigenvalues and Eigenvectors Only special vectors are eigenvectors - such vectors whose direction will not be changed by the transformation A (only scale) - they correspond to normal modes of the system act independently Examples eigenvalues 2, 3 eigenvectors Whatever A does to an arbitrary vector is fully determined by its eigenvalues and eigenvectors

23 Previously - Eigenvalues and Eigenvectors For the previous example We will get special solutions to ODE The linear combination is also a solution (due to the linearity of ) In the context of diff. equations special meaning Any solution can be expressed as linear combination Individual solutions correspond to modes

24 Eigenvalues of linear system Given linear system of differential equations ẋ = Ax For 2 dimesional system (A is 2 x 2), A has two eigenvalues Define = 1 2 and = if if < 0 > 0 > 0 saddle node we have two cases 1. eigenvalues positive < 0 2. eigenvalues negative : stable nodes of the system

25 Kinematic Position Control: Resulting Path 3.6.2

26 Linearization But our system is not linear, e.g. cannot be written in the form ẋ = Ax

27 Some terminology We have derived kinematics equations of the robot Non-linear differential equation In our case x = f 1 (x,y,θ,v,ω) x = f (x,u) y = f 2 (x,y,θ,v,ω) θ = f 3 (x, y,θ,v,ω)

28 Jacobian Matrix Suppose you have two dim function Gradient operator Jacobian is defined as F x = J F Linearization of a function F (x) =F (x 0 )+J F (x 0 )dx Linearization of system of diff. equations ẋ = J F (x 0 )dx + F (x 0 )

29 Kinematic Position Control: Stability Issue Continuous linear time-invariant system is exponentially stable if and only of the system has eigenvalues (i.e. poles of input-to-output systems) with strictly negative real parts Exponential Stability is a form of asymptotic stability In practice the system will not blow up give unbounded output, when given an finite input and non-zero initial condition

30 Kinematic Position Control: Stability Issue It can further be shown, that the closed loop control system is locally exponentially stable if > β α ρ Proof: linearize around equilibrium for small x -> cos x = 1, sin x = x k ρ 0 ; k < 0 ; k k > 0 and the characteristic polynomial of the matrix A of all roots have negative real parts.

31 Quadcopters model Popular unmanned areal vehicles (description adopted from (Robotics, Vision and control book, P. Corke T 3 T 4 ω 3 x ω 4 T 1 T 2 y ω 1 z ω 2 2 Upward thrust T i = bω i moving up in the negative z dir. Lift const. b depends on air density, blade radius and chord length

32 Quadcopters Translational dynamics (Newton s law includes mass/ acceleration/ forces) (Gravity Total thrust (rotated to the world frame)! m v = " 0 0 mg $! 0 R B % " Rotations are generated by pairwise differences in rotor thrusts (d distance from the center) Rolling and pitching torques around x and y Torque in z yaw torque Q i = kω i 2 τ z = (Q 1 Q 2 +Q 3 Q 4 ) 0 0 T $, T = T i i=1..4 % Torque applied by the motor as opposed to Aerodynamic drag τ x = dt 4 dt 2 τ x = db(ω 4 2 ω 2 2 ) τ y = db(ω 1 2 ω 3 2 )

33 Quadcopter dynamics! " Rotational Dynamics, rot. acceleration in the airframe, Euler s eq. of motion J ω = ω J ω + Γ, Where J is 3x3 inertia matrix Γ = [τ x,τ y,τ z ] T Forces and torques acting of the airframe obtained integrating forward the eq. above and Newton s second law (prev. slide) T τ x τ y τ z $! = % " b b b b 0 db 0 db db 0 db 0 k k k k The goal of control is then derive proper thrust and torque to achieve desired goal compute the rotor speeds! $ % " Substitute these to translational and rotational dynamics and get forward dynamics equations of quadropter ω 1 2 ω 2 2 ω 3 2 ω 4 2 $ %! " ω 1 2 ω 2 2 ω 3 2 ω 4 2 $! = A 1 % " T τ x τ y τ z $ %

34 Inertia Matrix Rotational Inertia of a body in 3D is represented by a 3x3 symmetric matrix J! J = " J xx J xy J xz J xy J yy J yz J xz J yz J zz $ % Diagonal elements are moments of inertia and offdiagonal are products of inertia Inertia matrix is a constant and depends on the mass and the shape of the body