Energy-based Swing-up of the Acrobot and Time-optimal Motion

Similar documents
Efficient Swing-up of the Acrobot Using Continuous Torque and Impulsive Braking

CONTROL OF THE NONHOLONOMIC INTEGRATOR

Stabilization of a 3D Rigid Pendulum

Nonlinear Tracking Control of Underactuated Surface Vessel

Robotics, Geometry and Control - A Preview

Control of the Inertia Wheel Pendulum by Bounded Torques

Swinging-Up and Stabilization Control Based on Natural Frequency for Pendulum Systems

A Normal Form for Energy Shaping: Application to the Furuta Pendulum

available online at CONTROL OF THE DOUBLE INVERTED PENDULUM ON A CART USING THE NATURAL MOTION

Passivity-based Stabilization of Non-Compact Sets

Stable Limit Cycle Generation for Underactuated Mechanical Systems, Application: Inertia Wheel Inverted Pendulum

A GLOBAL STABILIZATION STRATEGY FOR AN INVERTED PENDULUM. B. Srinivasan, P. Huguenin, K. Guemghar, and D. Bonvin

Angular Momentum Based Controller for Balancing an Inverted Double Pendulum

q HYBRID CONTROL FOR BALANCE 0.5 Position: q (radian) q Time: t (seconds) q1 err (radian)

Experimental Results for Almost Global Asymptotic and Locally Exponential Stabilization of the Natural Equilibria of a 3D Pendulum

Reverse Order Swing-up Control of Serial Double Inverted Pendulums

Stabilization and Passivity-Based Control

ENERGY BASED CONTROL OF A CLASS OF UNDERACTUATED. Mark W. Spong. Coordinated Science Laboratory, University of Illinois, 1308 West Main Street,

Stabilization of a Specified Equilibrium in the Inverted Equilibrium Manifold of the 3D Pendulum

1 The Observability Canonical Form

Control of the Underactuated Inertia Wheel Inverted Pendulum for Stable Limit Cycle Generation

A Novel Integral-Based Event Triggering Control for Linear Time-Invariant Systems

KINETIC ENERGY SHAPING IN THE INVERTED PENDULUM

Transverse Linearization for Controlled Mechanical Systems with Several Passive Degrees of Freedom (Application to Orbital Stabilization)

SWING UP A DOUBLE PENDULUM BY SIMPLE FEEDBACK CONTROL

Hybrid Control of the Pendubot

Weak Input-to-State Stability Properties for Navigation Function Based Controllers

EN Nonlinear Control and Planning in Robotics Lecture 3: Stability February 4, 2015

SWINGING UP A PENDULUM BY ENERGY CONTROL

A Fuzzy Control Strategy for Acrobots Combining Model-Free and Model-Based Control

Robust Adaptive Attitude Control of a Spacecraft

Acrobot stable walking in Hybrid systems notation*

El péndulo invertido: un banco de pruebas para el control no lineal. XXV Jornadas de Automática

Global stabilization of feedforward systems with exponentially unstable Jacobian linearization

Commun Nonlinear Sci Numer Simulat

Stability of Hybrid Control Systems Based on Time-State Control Forms

Average-Consensus of Multi-Agent Systems with Direct Topology Based on Event-Triggered Control

Lecture Note 7: Switching Stabilization via Control-Lyapunov Function

Swinging Up a Pendulum by Energy Control

Observer Design for a Flexible Robot Arm with a Tip Load

Topic # /31 Feedback Control Systems. Analysis of Nonlinear Systems Lyapunov Stability Analysis

Lecture 9 Nonlinear Control Design. Course Outline. Exact linearization: example [one-link robot] Exact Feedback Linearization

Case Study: The Pelican Prototype Robot

Trajectory Tracking Control of Bimodal Piecewise Affine Systems

Closed Loop Control of a Gravity-assisted Underactuated Snake Robot with Application to Aircraft Wing-Box Assembly

Minimizing Cable Swing in a Gantry Crane Using the IDA-PBC Methodology

Lyapunov-Based Controller for the Inverted Pendulum Cart System

ASTATISM IN NONLINEAR CONTROL SYSTEMS WITH APPLICATION TO ROBOTICS

Impulsive Stabilization and Application to a Population Growth Model*

Asymptotic Smooth Stabilization of the Inverted 3D Pendulum

Lecture 4. Chapter 4: Lyapunov Stability. Eugenio Schuster. Mechanical Engineering and Mechanics Lehigh University.

REMARKS ON THE TIME-OPTIMAL CONTROL OF A CLASS OF HAMILTONIAN SYSTEMS. Eduardo D. Sontag. SYCON - Rutgers Center for Systems and Control

High gain observer for embedded Acrobot

Lecture 9 Nonlinear Control Design

Minimum-Phase Property of Nonlinear Systems in Terms of a Dissipation Inequality

A NONLINEAR TRANSFORMATION APPROACH TO GLOBAL ADAPTIVE OUTPUT FEEDBACK CONTROL OF 3RD-ORDER UNCERTAIN NONLINEAR SYSTEMS

Exponential Controller for Robot Manipulators

Relaxed Matching for Stabilization of Mechanical Systems

Attitude Regulation About a Fixed Rotation Axis

Computing Optimized Nonlinear Sliding Surfaces

Research Article On the Stabilization of the Inverted-Cart Pendulum Using the Saturation Function Approach

TTK4150 Nonlinear Control Systems Solution 6 Part 2

Global stabilization of an underactuated autonomous underwater vehicle via logic-based switching 1

Underwater vehicles: a surprising non time-optimal path

q 1 F m d p q 2 Figure 1: An automated crane with the relevant kinematic and dynamic definitions.

Problem Description The problem we consider is stabilization of a single-input multiple-state system with simultaneous magnitude and rate saturations,

Rigid body stability and Poinsot s theorem

Robust Control of Cooperative Underactuated Manipulators

Constructive Invariant Manifolds to Stabilize Pendulum like systems Via Immersion and Invariance

Global output regulation through singularities

Model Orbit Robust Stabilization (MORS) of Pendubot with Application to Swing up Control

Applications of Controlled Invariance to the l 1 Optimal Control Problem

Convergence Rate of Nonlinear Switched Systems

An Explicit Characterization of Minimum Wheel-Rotation Paths for Differential-Drives

Control of Mobile Robots

Navigation and Obstacle Avoidance via Backstepping for Mechanical Systems with Drift in the Closed Loop

FINITE TIME CONTROL FOR ROBOT MANIPULATORS 1. Yiguang Hong Λ Yangsheng Xu ΛΛ Jie Huang ΛΛ

Throwing Motion Control of the Pendubot and Instability Analysis of the Zero Dynamics

Disturbance Attenuation for a Class of Nonlinear Systems by Output Feedback

Output Regulation of Uncertain Nonlinear Systems with Nonlinear Exosystems

Represent this system in terms of a block diagram consisting only of. g From Newton s law: 2 : θ sin θ 9 θ ` T

Modelling and Control of Mechanical Systems: A Geometric Approach

Hybrid active and semi-active control for pantograph-catenary system of high-speed train

IN this paper we consider the stabilization problem for

Stabilization of Nonlinear Systems via Forwarding

Dynamic Modeling of Rotary Double Inverted Pendulum Using Classical Mechanics

Rigid-Body Attitude Control USING ROTATION MATRICES FOR CONTINUOUS, SINGULARITY-FREE CONTROL LAWS

Nonholonomic Constraints Examples

Applied Math Qualifying Exam 11 October Instructions: Work 2 out of 3 problems in each of the 3 parts for a total of 6 problems.

Converse Lyapunov theorem and Input-to-State Stability

Nonlinear Systems and Control Lecture # 12 Converse Lyapunov Functions & Time Varying Systems. p. 1/1

An homotopy method for exact tracking of nonlinear nonminimum phase systems: the example of the spherical inverted pendulum

The Acrobot and Cart-Pole

Funnel control in mechatronics: An overview

Unit quaternion observer based attitude stabilization of a rigid spacecraft without velocity measurement

An introduction to Birkhoff normal form

Robust Semiglobal Nonlinear Output Regulation The case of systems in triangular form

IMECE NEW APPROACH OF TRACKING CONTROL FOR A CLASS OF NON-MINIMUM PHASE LINEAR SYSTEMS

Hybrid Systems Techniques for Convergence of Solutions to Switching Systems

Robust Control of Robot Manipulator by Model Based Disturbance Attenuation

Transcription:

Energy-based Swing-up of the Acrobot and Time-optimal Motion Ravi N. Banavar Systems and Control Engineering Indian Institute of Technology, Bombay Mumbai-476, India Email: banavar@ee.iitb.ac.in Telephone:(91)-(22) 2576 7888 Fax:(91)-(22) 2572 377 Arun D. Mahindrakar Systems and Control Engineering Indian Institute of Technology, Bombay Mumbai-476, India Email:arun@ee.iitb.ac.in Abstract We present a control law for the swing-up of an acrobot with torque constraints on the actuator. The domain of the initial condition for the strategy to work is the entire manifold and we further guarantee that the system reaches a small neighbourhood about the upward equilibrium position. For a restricted domain of the initial condition, we observe that the control law is bang-bang in nature. This motivates us to verify the time-optimality of the control strategy. Necessary conditions for time-optimality are presented and these are subsequently verified numerically. Index Terms nonholonomic systems, underactuated manipulator, energy-based control, time-optimal. I. INTRODUCTION Many control strategies have been presented for the acrobot [1], [2], [3], [4]. Most of them do not account for actuator saturation. Further, the domain of the initial condition for many of these are restrictive. Notions of time-optimality [5], [6], [7] of the acrobot motions have not received much attention either. In this paper we initially present a swing-up strategy for the acrobot that brings it to a small region around the upward equilibrium point. The domain of the initial condition for this control strategy is the entire manifold. We then verify this strategy for time-optimality based on a restricted domain of the initial condition. The paper is organised as follows. In section 2 we formulate the acrobot dynamics in a Hamiltonian framework. The Hamiltonian framework is advantageous in verifying timeoptimality. Section 3 presents the global control law that places the acrobot in a certain energy level. Section 4 presents the discussion on time-optimality. Section 5 numerically verifies the time-optimality and section 6 concludes the paper. II. HAMILTONIAN FORMULATION OF THE ACROBOT g lc1 link 2 lc2 q 1 DYNAMICS m1, m 2 = link masses l 1, l2 = link lengths I1, I 2 = link moments of inertia l c1,l c2 = centers of masses q 2 link 1 Fig. 1. Actuator The Acrobot For our purpose it is advantageous to use the Hamiltonian framework since it results in a constant control vector field [8] and this special structure proves useful in verifying the necessary conditions for the time-optimal problem that we formulate later. We write the equations of motion of the acrobot (schematic shown in Figure 1) defined on the configuration manifold Q = S 1 S 1 using a Hamiltonian formulation. The configuration space Q is parametrized by the joint angles (q 1, q 2 ) and the generalized momentum is defined

as p = D(q) q, where D(q) is the inertia matrix defined as [ ] (c 1 + c 2 + 2c 3 cos q 2 ) (c 2 + c 3 cos q 2 ) D(q) =. (c 2 + c 3 cos q 2 ) c 2 The inertial parameters are collected in the following constants c i, i = 1,..., 5 as c 1 = m 1 l 2 c1 + m 2 l 2 1 + I 1, c 2 = m 2 l 2 c2 + I 2, c 3 = m 2 l 1 l c2, c 4 = m 1 l c1 + m 2 l 1, c 5 = m 2 l c2. With the state vector defined as z = (z 1 = (q 1 π/2), z 2 = q 2, z 3 = p 1, z 4 = p 2 ), the Hamiltonian system is given by ż = f(z) + bu, (1) where the state space manifold is M = S 1 S 1 IR 2 and the smooth vector fields f and b are given by [ ] D 1 z 3 z 4 f(z) = V 1 [ ] ; b = 1 2 [z z 3 z 4 ] D 1 3 2 V 2 1 with V = c 4 g cos z 1 + c 5 g cos(z 1 + z 2 ) and u = τ 2 Ω, the class of admissible controls defined as z 4 Ω = {u IR : u β, β > }. The equilibrium solutions of (1) with the input equal to zero constitute an important class of solutions. The set of equilibrium solution z e corresponding to u = is given by {z e M : z e 1 = k 1 π, z e 2 = k 2 π, z e 3 = z e 4 = }, k 1, k 2 =, 1. III. A GLOBAL DISCONTINUOUS CONTROL LAW The control objective is to bring the acrobot to a desired energy level that corresponds to that at the upward equilibrium point. So the first objective is to pump in the requisite energy and then guarantee that the system reaches a small neighbourhood about the upward equilibrium point. The latter fact is proved using Birkhoff s theorem on Ω-limit points [9]. Denote the energy of the acrobot at the four equilibrium positions by E dd, E du, E ud, E uu where the first subscript denotes the position of the 1st link and the second subscript denotes the position of the second link; the subscript u denotes upright and the subscript d denotes downright position of the link. Let where E uu = (c 4 + c 5 )g. Ê(z) = E(z) E uu Theorem 1: Given the torque constraint u(t) β the control law { δ (δ is small) if z B u = β sign[ê(z) χ(z)] otherwise (2) where B = {( π,,, ), ( π, π,, ), (, π,, )} and χ(z) = (c 1 + c 2 + 2c 3 cos z 2 )z 4 (c 2 + c 3 cos z 2 )z 3 (c 2 c 1 c 2 3 cos2 z 2 ) moves the acrobot to an energy level Ê(z) =. Proof: Note that the positive definiteness of the inertia matrix ensures (c 1 c 2 c 2 3 cos z 2 2 ) > z 2 S 1 and so χ(z) is defined for all z M. Further, χ(z) = when z 3 = z 4 = and χ(z) only at the equilibrium points. Now let us examine the dynamics of Ê for which we consider a candidate Lyapunov function V 2 : M IR defined by We have V 2 (z) = 1 2Ê2 (z) (3) V 2 (z) = Ê(z) Ê (4) and using the passivity property of the acrobot, (4) becomes V 2 = Ê(z)χ(z)u (5) Substituting the control law (2) results in { if z B V 2 = β Ê(z) χ(z) otherwise Since V 2 is non-increasing, the trajectory is bounded and the solution of the closed loop system remains inside a compact set defined by Ω c = {z M : V 2 (z) V 2 (z())}. Let Q = {z Ω c : V 2 = }. The set Q is given by Q = {(z M : χ(z) = ) (z M : Ê(z) = )}. Let M be the largest invariant set in Q. Now we compute M. Suppose Ê(z) and χ(z) =. We have two cases. 1) χ(z) = and z B. In this case the small control δ perturbs the system out of these equilibria. 2) χ(z) = and z / B. The dynamics of the system ensures that the system moves out of such points. Hence M = {z Q : Ê(z) = } is the largest invariant set. Remark 1: Note that the control law given by (2) reduces to u = βsign[ê(z)χ(z)] (6)

if the initial condition satisfies Ê(z()) max{(e dd E uu ), (E du E uu ), (E ud E uu )}. This condition would ensure that the acrobot does not get stuck at any of the intermediate equilibrium positions. In practice we would like to capture the acrobot in a region close to its upward equilibrium position. Let ˆT > be the instant at which Ê(z) = ɛ 1 where ɛ 1 > is sufficiently small. We let the control u(t) =, t > ˆT and thus the state z evolves on the set defined by Π = {z M : E(z) = E uu ɛ 1 }. (7) Lemma 3.1: Π is the one and only one non-empty, closed and invariant set in Π. Proof : We return to the Lagrangian formulation for this proof. Let (x 1, x 2, x 3, x 4 ) be the state variables and let K be any arbitrary, non-empty, closed set in Π. Then K is of the form K = {[a 1, a 2 ] [b 1, b 2 ] Π} where a 1, a 2, b 1, b 2 [, 2π). Note that once the link angles (a p, b p ) are specified, the link velocities (d p, e p ) must satisfy the equation where V (a p, b p ) + T (b p, d p, e p ) = E uu ɛ 1 (8) V (a p, b p ) = c 4 g cos a p + c 5 g cos(a p + b p ) T (d p, e p ) = d2 p 2 (c 1 + c 2 + 2c 3 cos b p ) +d p e p (c 2 + c 3 cos b p ) + c 2e 2 p 2 Notice that (8) is quadratic in both d p and e p. Define κ 1 = c 2 /2 > κ 2 = d p (c 2 + c 3 cos b p ) κ 3 = E uu ɛ 1 d2 p 2 (c 1 + c 2 + 2c 3 cos b p ) c 4 g cos a p c 5 g cos(a p + b p ) and rewrite the quadratic equation as A permissible d p satisfies κ 3 + κ 2 e p + κ 1 e 2 p = κ 2 2 + 4κ 1 κ 3 and there exist two value of e p given by ẋ 2 (ap,b p,d p) = κ 2 ± κ 2 2 + 4κ 1κ 3 2κ 1 (9) Now consider the left-extreme link angles (a 1, b 1 ) K and an admissible d 1. Two cases are possible κ 3. Then one of the solutions is e 1 < and the trajectory originating at (a 1, b 1, d 1, e 1 ) K leaves K. κ 3 =. Then one of the solutions is e 1 = and the other κ 2 κ 1. If κ 2 > then once again the trajectory originating at (a 1, b 1, d 1, e 1 ) K leaves K. If κ 2, we cannot reach any conclusion and we then consider a pair of link angles (a p, b 2 ) such that a p > a 1 and κ 3. Note that such an a p exists since κ 3 is not a constant function with respect to a p. Then one of the values of e p < and the trajectory originating at (a p, b 1, d p, e p ) K leaves K. Hence the set K is not invariant. But K is any arbitrary, nonempty, closed set in Π. Hence Π is the only invariant set. Definition 3.1: Let y (t) denote an integral curve of a system ẏ = G(y) which is assumed to be defined for all t <. A point ȳ is said to be a ω-limit point of y o (t) if there exists an increasing sequence of values of t such that t 1 < t 2 < t 3... < t k, lim k t k =, lim k yo (t k ) = ȳ. The set Ω o of all ω-limit points of y o (t) is the ω-limit set of y o (t). Theorem 2: [G. D. Birkhoff] Suppose y o (t) is a bounded trajectory. Its ω-limit set Ω o is nonempty, closed and invariant under the flow φ t G. Theorem 3: Let ξ = (a exti, b exti,, ) Π. Then ξ is an ω-limit point of any trajectory that begins in the set Π. Proof: Any trajectory that originates in the set Π is bounded. From Birkhoff s theorem [9], the Ω-limit set of any such bounded trajectory is nonempty, closed and invariant. From lemma 3.1, the only set that satisfies these properties is Π. It follows that ξ is an ω-limit point. Remark 2: Note that for ɛ 1 =, the upward equilibrium point is an ω-limit point of any trajectory that originates in Π. Once the system is in the neighbourhood of the upward equilibrium point, a linear feedback control can be switched on to balance the acrobot at the desired configuration. In the following section we investigate the time-optimality of the proposed control. IV. TIME-OPTIMAL CONTROL Consider the energy-pumping control law as in (6) with a restriction in the domain of the initial condition. The

bang-bang nature of the law motivates us to pose the question Is the control (6) optimal in a sense that it achieves energypumping in minimum time? While obtaining necessary and sufficient conditions for this problem is not trivial, we suggest a procedure to satisfy the necessary conditions for time-optimality. Time optimal problem: Minimize the performance measure J(t f ) = t f (1) for the system (1) with the initial condition {z() M : Ê(z()) max{(e dd E uu ), (E du E uu ), (E ud E uu )}} and the constraint that the final state lies on the surface S(z(t f )) = where S : M IR is defined by The control belongs to the set S(z) = E(z) E uu + ɛ 1. (11) {u( ) : IR IR : u(t) β} Writing the Hamiltonian for the above problem as H(z, u, λ) = λ T [f(z) + bu] (12) where λ(t) R 4 is the co-state vector and denoting the optimal trajectories and control law by, from Pontryagin s minimum principle (PMP) the optimal control law satisfies H(z, u, λ ) H(z, u, λ ) for all admissible u. The above inequality leads to a bang-bang control law of the form u = β sign [ λ T b ]. (13) The necessary conditions for optimality are 1) z and λ are the solutions of the canonical equations ż = H λ (z, u, λ ) (14) = f(z ) + bu [ f λ = + b ] T with the boundary conditions z () = z. (z=z ) λ (15) 2) The variation δz f should be such that it satisfies the transversality condition [1] S(z) T δz f = (16) (z=z (t f )) and λ (t f ) T δz f =. (17) 3) The Hamiltonian at the final time t f is H(z (t f ), u (t f ), λ (t f )) = 1 (18) Theorem 4: Let z b be the trajectory generated by the control law given by (6). Suppose the switching times are t s1,..., t sp. Let Φ(t, t f ) be the state-transition matrix of the equation ( f(z) ) T ψ = ψ, ψ(t) R 4 t [, t f ] (z=z b ) A necessary condition for time-optimality of the trajectory z b is that the vector [α 1 α 2 α 3 1] T where ( S α i = / S ), i = 1, 2, 3. i 4 (z=z b (t f )) is orthogonal to the subspace spanned by the vectors {Φ(t si, t f ) T b}. S Proof: We firts note that 4 = χ(z) and χ(z b (t f )). Therefore α i z=zb (t f ), i = 1, 2, 3 are well-defined. Now, equations (16) and (17) can be recast into a single one as where λ(t f ) T Q(z b (t f )) Q(z b (t f )) = δz 1f δz 2f δz 3f = 1 1 1 α 1 α 2 α 3. (19) Since [δz 1f δz 2f δz 3f ] T is arbitrary, its coefficient must be zero. Accordingly, we have Q T (z b (t f ))λ(t f ) = or λ(t f ) N (Q T (z b (t f )) where N denotes the null space. The non-trivial solution of the above equation is of the form λ(t f ) = α 1 α 2 α 3 1 λ 4(t f ) λ 4 (t f ) (2)

In view of the constant control vector field, the costate equation (15) becomes ( f(z) ) T λ = λ. (21) ( ) Denote A(z b ) = T f the form (z=z b ) (z=z b ). The solution to (21) is of λ(t) = Φ(t, t f )λ(t f ) (22) where Φ(t, t f ) IR 4 4 is the state transition matrix, which is nonsingular and satisfies Φ(t, t f ) = A(z b )Φ(t, t f ), Φ(t f, t f ) = I. (23) Suppose the switching times are t s1,..., t sp. Then from the necessary condition (13) we have or b T λ(t si ) = b T Φ(t si, t f )λ(t f ) = for each i = 1,..., p (24) < Φ(t si, t f ) T b, [α 1 α 2 α 3 1] T >= for each i = 1,..., p Remark 3: Note that the Hamiltonian at the final time is λ T (t f )(f(z b (t f )) + bu(t f )) = 1 Now < f(z b (t f )), λ(t f ) >= since [ ] I 2 2 S f(z) = I 2 2. and the condition on the Hamiltonian reduces to u(t f ) < b, λ(t f ) >= 1 Hence λ 4 (t f ) = 1 u(t f ) where u(t f ) = ±β. We next present the numerical verification of these results. V. NUMERICAL VERIFICATION OF PONTRYAGIN S MINIMUM PRINCIPLE To verify the necessary condition for the time-optimality of the proposed control law (6), we follow these steps. 1) Fix β and the initial condition z(). 2) Apply the control (6) to the system till the instant Ê = ɛ 1. Denote that instant as t f. Store the resulting time histories of the state z b from to t f. Define s(t) = Ê(z b )χ(z b ). 3) Compute α 1, α 2, α 3. 4) Numerically solve (23) over the interval [, t f ] to obtain Φ(t, t f ). The costate is obtained as λ(t) = Φ(t, t f )λ(t f ), t [, t f ]. 5) The generated switching function is given by s a (t) =< b, λ(t) >= λ 4 (t). 6) Verify the orthogonality condition (24 ). This requires that the switching functions s and s a match. By matching it is meant that they satisfy s(t si ) = s a (t si ) =, i = 1,..., p. A. Simulation results The acrobot parameters used in the simulations are l 1 = 1 m, l 2 = 2 m, m 1 = 1 kg, m 2 = 2 kg, I 1 =.83 kg m 2, I 2 =.667 kg m 2. We let β = 4 and the initial vector is z() = ( π,,.5,.5). With these we have one switching (p = 1). The control u(t) is switched off when Ê =.1. The final time is t f = 4.664 seconds. The costate vector at the final time t f is given by λ(t f ) = (.9993,.585,.58,.25). We repeat the steps (1-6) outlined in section V for different initial conditions (see Table I). We have the following observations. Numerical results TABLE I EFFECT OF THE NUMBER OF SWITCHINGS p ON THE MATCHING OF SWITCHING FUNCTIONS Case z() p s and s a match? 1 ( π,,.5,.5) 1 Yes 2 ( π,,, 1) 3 Yes 3 ( π, π, 1, 1) 5 No indicate that the number of switchings p influences the timeoptimality property. In particular, it is seen that for p 3, the control law (6) satisfies the necessary conditions for optimality. Joint angles (rad) Lyapunov function V 1 5 5 1 2 4 6 8 1 4 Torque (Nm) 2 2 4 2 4 6 8 1 15 1 5 z 1 z 2 2 4 6 8 1 Fig. 2. Angular momentum kg m 2 s 1 Energy profile Switching function s 2 2 z 3 z 4 4 2 4 6 8 1 6 E d 4 2 2 2 4 6 8 1 5 5 1 2 4 6 8 1 Energy build-up phase

VI. CONCLUSIONS A global discontinuous control law has been presented for the acrobot and it guarantees that the system reaches a small neighbourhood around the upward equilibrium point of interest. For a certain domain of the initial conditions that is based on an energy requirement, necessary conditions for time-optimality of the control law are presented. ACKNOWLEDGEMENT This work was supported by the Department of Science and Technology (DST) as a Sponsored Research & Development project (Sanction No. 1/IFD/28/21-2). REFERENCES [1] M. W. Spong, The swing up control problem for the acrobot, IEEE Control Systems Magazine, vol. 15, pp. 49 55, February 1995. [2] K. J. Åström and K. Furuta, Swinging up a pendulum by energy control, in Proceedings of the 13 th IFAC World Congress, vol. E, (San Francisco), pp. 37 42, 1996. [3] A. S. Shiriaev, H. Ludvigsen, O. Egeland, and A. L. Fradkov, Swinging up of non-affine in control pendulum, in Proceedings of the American Control Conference, (San Diego, California), pp. 439 444, 199. [4] I. Fantoni, R. Lozano, and M. W. Spong, Energy based control of the pendubot, IEEE Trans. on Automatic Control, vol. 45, pp. 725 729, April 2. [5] K. G. Shin and N. D. McKay, Minimum-time control of robotic manipulators with geometric path constraints, IEEE Trans. on Automatic Control, vol. 3, pp. 531 541, June 1985. [6] E. D. Sontag and H. J. Sussmann, Time-optimal control of manipulators, in Proceedings of IEEE International Conference on Robotics & Automation, pp. 1692 1697, 1986. [7] G. Sahar and J. M. Hollerbach, Planning of minimum-time trajectories for robot-arms, in Proceedings of the IEEE International Conference on Robotics and Automation, (St. Louis, Missouri), pp. 751 758, 1985. [8] Y. Chen and A. A. Descrochers, A Proof of the structure of the minimum-time control law of robotic manipulator using a Hamiltonian formulation, IEEE Trans. on Robotics & Automation, vol. 6, pp. 388 393, June 199. [9] A. Isidori, Nonlinear Control Systems. New York: Spinger Verlag, 1995. [1] F. L. Lewis and V. L. Syrmos, Optimal Control. New York: John Wiley & Sons, Inc., 1995.