numerical optimal control with daes lecture 8 direct
play

Numerical Optimal Control with DAEs Lecture 8: Direct Collocation S - PowerPoint PPT Presentation

Oct 20, 2022 •2.42k likes •3.7k views

Numerical Optimal Control with DAEs Lecture 8: Direct Collocation S ebastien Gros AWESCO PhD course 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 1 / 24 Survival map of Direct Optimal Control Collocation

  1. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  2. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  3. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  4. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  5. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  6. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  7. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  8. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  9. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  10. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  11. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i ���� � �� � i =0 parameters polynomials having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  12. E.g. Polynomial interpolation t k = 1 , t k +1 = 2 Consider a time grid : K = 4 { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] { t k , 0 , ..., t k , K } = { 1 . 0 , 1 . 0694 , 1 . 33 , 1 . 67 , 1 . 931 } Lagrange Polynomials : 1.5 t k, 0 t k, 1 t k, 2 t k, 3 t k, 4 K � t − t k , j P k , i ( t ) = ∈ R 1 t k , i − t k , j j =0 , j � = i 0.5 of order K , with property: � 1 if l = i 0 P k , i ( t k , l ) = 0 if l � = i t k t k +1 -0.5 Interpolation with θ k , i ∈ R n 1 1.2 1.4 1.6 1.8 2 K � x ( θ k , t ) = · P k , i ( t ) θ k , i Note: the Lagrange polynomials are orthogonal, i.e. ���� � �� � � t k +1 i =0 parameters polynomials P k , i ( t ) P k , j ( t ) d t = 0 , ∀ i � = j t k having the property: x ( θ , t k , j ) = θ k , j 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 5 / 24

  13. Outline Polynomial interpolation 1 Collocation-based integration 2 Collocation in multiple-shooting 3 Direct Collocation 4 NLP from direct collocation 5 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 6 / 24

  14. Collocation methods - key idea Approximate state trajectory x ( t ) via polynomials (order K ) t 0 t 1 t 2 t 3 t 4 t 5 0.35 f ( x 4 , u 4 ) x 4 0.3 x 3 0.25 f ( x 3 , u 3 ) 0.2 State f ( x 2 , u 2 ) x 2 0.15 0.1 x 1 f ( x 1 , u 1 ) 0.05 x 0 0 f ( x 0 , u 0 ) -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 7 / 24

  15. Collocation methods - key idea Approximate state trajectory x ( t ) via polynomials (order K ) t k t k +1 3 f ( x k , u k ) 2 x k +1 1 0 -1 x k -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 7 / 24

  16. Collocation methods - key idea Approximate state trajectory x ( t ) via polynomials (order K ) Time grid : { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] t k t k +1 3 f ( x k , u k ) 2 x k +1 1 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 7 / 24

  17. Collocation methods - key idea Approximate state trajectory x ( t ) via polynomials (order K ) Time grid : { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] Interpolate on each interval [ t k , t k +1 ] using: K � x ( θ k , t ) = θ k , i · P k , i ( t ) ���� � �� � i =0 parameters polynomials t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 7 / 24

  18. Collocation methods - key idea Approximate state trajectory x ( t ) via polynomials (order K ) Time grid : { t k , 0 , ..., t k , K } ∈ [ t k , t k +1 ] Interpolate on each interval [ t k , t k +1 ] using: K � x ( θ k , t ) = θ k , i · P k , i ( t ) ���� � �� � i =0 parameters polynomials Integration: adjust θ k , i to approximate the dynamics ˙ x = F ( x , u ) t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 7 / 24

  19. Collocation methods - how to adjust the θ k , i ? On each interval [ t k , t k +1 ], approximate ˙ x = F ( x , u k ) using K � x ( θ k , t ) = · P k , i ( t ) with x ( θ k , t k , j ) = θ k , j θ k , i ���� � �� � i =0 parameters polynomials Note: we have K + 1 degrees of freedom per state . t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 8 / 24

  20. Collocation methods - how to adjust the θ k , i ? On each interval [ t k , t k +1 ], approximate ˙ x = F ( x , u k ) using K � x ( θ k , t ) = · P k , i ( t ) with x ( θ k , t k , j ) = θ k , j θ k , i ���� � �� � i =0 parameters polynomials Note: we have K + 1 degrees of freedom per state . Collocation uses the constraints: x ( θ k , t k ) = θ k , 0 = x k t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 8 / 24

  21. Collocation methods - how to adjust the θ k , i ? On each interval [ t k , t k +1 ], approximate ˙ x = F ( x , u k ) using K � x ( θ k , t ) = · P k , i ( t ) with x ( θ k , t k , j ) = θ k , j θ k , i ���� � �� � i =0 parameters polynomials Note: we have K + 1 degrees of freedom per state . Collocation uses the constraints: x ( θ k , t k ) = θ k , 0 = x k ∂ ∂ t x ( θ k , t k , j ) = F ( x ( θ k , t k , j ) , u k ) , j = 1 , ..., K t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 8 / 24

  22. Collocation methods - how to adjust the θ k , i ? On each interval [ t k , t k +1 ], approximate ˙ x = F ( x , u k ) using K � x ( θ k , t ) = · P k , i ( t ) with x ( θ k , t k , j ) = θ k , j θ k , i ���� � �� � i =0 parameters polynomials Note: we have K + 1 degrees of freedom per state . Collocation uses the constraints: x ( θ k , t k ) = θ k , 0 = x k (note that x k , u k are coming from the NLP !!) ∂ ∂ t x ( θ k , t k , j ) = F ( x ( θ k , t k , j ) , u k ) , j = 1 , ..., K t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 8 / 24

  23. Collocation methods - how to adjust the θ k , i ? On each interval [ t k , t k +1 ], approximate ˙ x = F ( x , u k ) using K � x ( θ k , t ) = · P k , i ( t ) with x ( θ k , t k , j ) = θ k , j θ k , i ���� � �� � i =0 parameters polynomials Note: we have K + 1 degrees of freedom per state . Collocation uses the constraints: x ( θ k , t k ) = θ k , 0 = x k (note that x k , u k are coming from the NLP !!) ∂ ∂ t x ( θ k , t k , j ) = F ( θ k , j , u k ) , j = 1 , ..., K t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 8 / 24

  24. Collocation methods - how to adjust the θ k , i ? On each interval [ t k , t k +1 ], approximate ˙ x = F ( x , u k ) using K � x ( θ k , t ) = · P k , i ( t ) with x ( θ k , t k , j ) = θ k , j θ k , i ���� � �� � i =0 parameters polynomials Note: we have K + 1 degrees of freedom per state . Collocation uses the constraints: x ( θ k , t k ) = θ k , 0 = x k (note that x k , u k are coming from the NLP !!) K � θ k , i · ˙ P k , i ( t k , j ) = F ( θ k , j , u k ) , j = 1 , ..., K i =0 t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 8 / 24

  25. Collocation methods - Implementation Collocation uses the constraints: θ k , 0 = x k K � θ k , i · ˙ P k , i ( t k , j ) = F ( θ k , j , u k ) i =0 for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  26. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  27. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  28. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  29. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  30. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  31. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  32. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  33. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  34. Collocation methods - Implementation Solve for θ k , i using Newton Collocation uses the constraints:   θ k , 0 = x k θ k , 0 − x k � K i =0 θ k , i ˙ P k , i ( t k , 1 ) − F ( θ k , 1 , u k ) K   �    = 0 θ k , i · ˙ . P k , i ( t k , j ) = F ( θ k , j , u k )   . .  i =0 � K i =0 θ k , i ˙ P k , i ( t k , K ) − F ( θ k , K , u k ) for j = 1 , ..., K . End-state: K � x ( θ k , t k +1 ) = θ k , i · P k , i ( t k +1 ) i =0 t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  35. Collocation methods - Implementation Shooting constraints Collocation uses the constraints: θ k , 0 = x k f ( x k , u k ) − x k +1 = 0 � �� � K � = x ( θ k , t k +1 ) θ k , i · ˙ P k , i ( t k , j ) = F ( θ k , j , u k ) becomes: i =0 K for j = 1 , ..., K . End-state: � θ k , i P k , i ( t k +1 ) − x k +1 = 0 K i =0 � x ( θ k , t k +1 ) = θ k , i · P k , i ( t k +1 ) i =0 t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 9 / 24

  36. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  37. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) Collocation points on [0 , 1] : 2 θ k, 2 x k +1 1 x ( θ k , t ) K Legendre Radau 0 θ k, 1 1 0 . 5 1 . 0 θ k, 0 -1 x k 0.211325 0.333333 t k, 0 t k, 1 t k, 2 t k, 3 2 -2 0.788675 1.000000 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.112702 0.155051 3 0.500000 0.644949 0.887298 1.000000 0.069432 0.088588 0.330009 0.409467 4 0.669991 0.787659 0.930568 1.000000 0.046910 0.057104 0.230765 0.276843 5 0.500000 0.583590 0.769235 0.860240 0.953090 1.000000 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  38. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) Collocation points on [0 , 1] : 2 θ k, 2 x k +1 1 x ( θ k , t ) K Legendre Radau 0 θ k, 1 1 0 . 5 1 . 0 θ k, 0 -1 x k 0.211325 0.333333 t k, 0 t k, 1 t k, 2 t k, 3 2 -2 0.788675 1.000000 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.112702 0.155051 3 0.500000 0.644949 0.887298 1.000000 0.069432 0.088588 0.330009 0.409467 4 0.669991 0.787659 0.930568 1.000000 0.046910 0.057104 0.230765 0.276843 5 0.500000 0.583590 0.769235 0.860240 0.953090 1.000000 +one point at t k to enforce continuity ! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  39. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) Collocation points on [0 , 1] : 2 θ k, 2 x k +1 1 x ( θ k , t ) K Legendre Radau 0 θ k, 1 1 0 . 5 1 . 0 θ k, 0 -1 x k 0.211325 0.333333 t k, 0 t k, 1 t k, 2 t k, 3 2 -2 0.788675 1.000000 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.112702 0.155051 Why these points ?!? They deliver an 3 0.500000 0.644949 exact integration for any polynomial P of 0.887298 1.000000 order < 2 K (Legendre) and < 2 K − 1 0.069432 0.088588 (Radau). I.e. for 0.330009 0.409467 4 0.669991 0.787659 x = F ( x , u ) = P ( t ) ˙ 0.930568 1.000000 0.046910 0.057104 the collocation equations deliver an exact 0.230765 0.276843 solution, namely: 5 0.500000 0.583590 � t k +1 0.769235 0.860240 x ( t k +1 , θ k ) = x k + P ( τ ) d τ 0.953090 1.000000 t k +one point at t k to enforce continuity ! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  40. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) Collocation points on [0 , 1] : 2 θ k, 2 x k +1 1 x ( θ k , t ) K Legendre Radau 0 θ k, 1 1 0 . 5 1 . 0 θ k, 0 -1 x k 0.211325 0.333333 t k, 0 t k, 1 t k, 2 t k, 3 2 -2 0.788675 1.000000 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.112702 0.155051 Interval [ t k , t k +1 ] ?? 3 0.500000 0.644949 0.887298 1.000000 Rescale & translate the collocation 0.069432 0.088588 points to [ t k , t k +1 ] 0.330009 0.409467 4 0.669991 0.787659 0.930568 1.000000 0.046910 0.057104 0.230765 0.276843 5 0.500000 0.583590 0.769235 0.860240 0.953090 1.000000 +one point at t k to enforce continuity ! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  41. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) Collocation points on [0 , 1] : 2 θ k, 2 x k +1 1 x ( θ k , t ) K Legendre Radau 0 θ k, 1 1 0 . 5 1 . 0 θ k, 0 -1 x k 0.211325 0.333333 t k, 0 t k, 1 t k, 2 t k, 3 2 -2 0.788675 1.000000 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.112702 0.155051 Interval [ t k , t k +1 ] ?? 3 0.500000 0.644949 0.887298 1.000000 Rescale & translate the collocation 0.069432 0.088588 points to [ t k , t k +1 ], or... 0.330009 0.409467 4 Modification of the collocation 0.669991 0.787659 equations with h k = t k +1 − t k : 0.930568 1.000000 K � 0.046910 0.057104 θ k , j ˙ P k , j ( t k , i ) = h k F ( θ k , i , u k ) 0.230765 0.276843 j =0 5 0.500000 0.583590 0.769235 0.860240 0.953090 1.000000 +one point at t k to enforce continuity ! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  42. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) Collocation points on [0 , 1] : 2 θ k, 2 x k +1 1 x ( θ k , t ) K Legendre Radau 0 θ k, 1 1 0 . 5 1 . 0 θ k, 0 -1 x k 0.211325 0.333333 t k, 0 t k, 1 t k, 2 t k, 3 2 -2 0.788675 1.000000 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.112702 0.155051 Interval [ t k , t k +1 ] ?? 3 0.500000 0.644949 0.887298 1.000000 Rescale & translate the collocation 0.069432 0.088588 points to [ t k , t k +1 ], or... 0.330009 0.409467 4 Modification of the collocation 0.669991 0.787659 equations with h k = t k +1 − t k : 0.930568 1.000000 K � 0.046910 0.057104 θ k , j ˙ P k , j ( t k , i ) = h k F ( θ k , i , u k ) 0.230765 0.276843 j =0 5 0.500000 0.583590 Careful if F is time-dependent ! 0.769235 0.860240 0.953090 1.000000 +one point at t k to enforce continuity ! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  43. Selection of the time grid t k , i t k t k +1 θ k, 3 3 f ( x k , u k ) Collocation points on [0 , 1] : 2 θ k, 2 x k +1 1 x ( θ k , t ) K Legendre Radau 0 θ k, 1 1 0 . 5 1 . 0 θ k, 0 -1 x k 0.211325 0.333333 t k, 0 t k, 1 t k, 2 t k, 3 2 -2 0.788675 1.000000 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.112702 0.155051 Interval [ t k , t k +1 ] ?? 3 0.500000 0.644949 0.887298 1.000000 Rescale & translate the collocation 0.069432 0.088588 points to [ t k , t k +1 ], or... 0.330009 0.409467 4 Modification of the collocation 0.669991 0.787659 equations with h k = t k +1 − t k : 0.930568 1.000000 K � 0.046910 0.057104 θ k , j ˙ P k , j ( t k , i ) = h k F ( θ k , i , u k ) 0.230765 0.276843 j =0 5 0.500000 0.583590 Careful if F is time-dependent ! 0.769235 0.860240 0.953090 1.000000 Note that Radau has a collocation point at the end of the interval, i.e. θ k , K provides +one point at t k to enforce continuity ! the end-state of the integration ! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 10 / 24

  44. Stability & Order Collocation methods are A-stable (i.e. can handle stiff equations). They have no stability limitation on the time intervals h = t k +1 − t k for stiff problems. I.e. even large time steps h = t k +1 − t k allow for capturing steady state and slow dynamics. Radau collocation is additionally L-stable . I.e. it can handle eigenvalues at −∞ . On an interval h k = t k +1 − t k , the integration error is O ( h 2 K k ) for Legendre and O ( h 2 K − 1 ) for Radau. Losing one order is the ”price” for having a collocation point k at t k +1 . The integration error applies to the end-state of the integrator, but not to the intermediate points ! Collocation-based integration is an Implicit Runge-Kutta scheme. Implicit Euler is an order-1 scheme ! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 11 / 24

  45. Collocation - Sensitivity Collocation constraints... θ k , 0 = x k K � , i = 1 , ..., K � P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  46. Collocation - Sensitivity Collocation constraints... ... can be seen as θ k , 0 = x k c ( x k , u k , θ k ) = 0 K � , i = 1 , ..., K � P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  47. Collocation - Sensitivity Collocation constraints... ... can be seen as θ k , 0 = x k c ( x k , u k , θ k ) = 0 K � , i = 1 , ..., K � Solved by iterating: P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 − 1 ∆ θ k = − ∂ c ( x k , u k , θ k ) c ( x k , u k , θ k ) ∂ θ k 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  48. Collocation - Sensitivity Collocation constraints... ... can be seen as θ k , 0 = x k c ( x k , u k , θ k ) = 0 K � , i = 1 , ..., K � Solved by iterating: P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 − 1 ∆ θ k = − ∂ c ( x k , u k , θ k ) c ( x k , u k , θ k ) ∂ θ k Integrator function given by: f ( x k , u k ) = x ( θ k , t k +1 ) 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  49. Collocation - Sensitivity Collocation constraints... ... can be seen as θ k , 0 = x k c ( x k , u k , θ k ) = 0 K � , i = 1 , ..., K � Solved by iterating: P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 − 1 ∆ θ k = − ∂ c ( x k , u k , θ k ) c ( x k , u k , θ k ) ∂ θ k Integrator function given by: f ( x k , u k ) = x ( θ k , t k +1 ) Get sensitivities using: ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ θ k ∂ θ k , ∂ x k ∂ θ k ∂ x k ∂ u k ∂ θ k ∂ u k 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  50. Collocation - Sensitivity Collocation constraints... ... can be seen as θ k , 0 = x k c ( x k , u k , θ k ) = 0 K � , i = 1 , ..., K � Solved by iterating: P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 − 1 ∆ θ k = − ∂ c ( x k , u k , θ k ) c ( x k , u k , θ k ) ∂ θ k Integrator function given by: f ( x k , u k ) = x ( θ k , t k +1 ) Get sensitivities using: ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ θ k ∂ θ k , ∂ x k ∂ θ k ∂ x k ∂ u k ∂ θ k ∂ u k Implicit function theorem states that ∂ c ∂ θ k + ∂ c ∂ c ∂ θ k + ∂ c = 0 , = 0 ∂ θ k ∂ x k ∂ x k ∂ θ k ∂ u k ∂ u k 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  51. Collocation - Sensitivity Collocation constraints... ... can be seen as θ k , 0 = x k c ( x k , u k , θ k ) = 0 K � , i = 1 , ..., K � Solved by iterating: P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 − 1 ∆ θ k = − ∂ c ( x k , u k , θ k ) c ( x k , u k , θ k ) ∂ θ k Integrator function given by: f ( x k , u k ) = x ( θ k , t k +1 ) Get sensitivities using: ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ θ k ∂ θ k , ∂ x k ∂ θ k ∂ x k ∂ u k ∂ θ k ∂ u k Implicit function theorem states that ∂ c ∂ θ k + ∂ c ∂ c ∂ θ k + ∂ c = 0 , = 0 ∂ θ k ∂ x k ∂ x k ∂ θ k ∂ u k ∂ u k Hence: − 1 ∂ c − 1 ∂ c ∂ θ k = − ∂ c ∂ θ k = − ∂ c , ∂ x k ∂ θ k ∂ x k ∂ u k ∂ θ k ∂ u k 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  52. Collocation - Sensitivity Collocation constraints... ... can be seen as θ k , 0 = x k c ( x k , u k , θ k ) = 0 K � , i = 1 , ..., K � Solved by iterating: P k , j ( t k , i ) = F � θ k , i , u k θ k , j ˙ j =0 − 1 ∆ θ k = − ∂ c ( x k , u k , θ k ) c ( x k , u k , θ k ) ∂ θ k Integrator function given by: f ( x k , u k ) = x ( θ k , t k +1 ) Get sensitivities using: ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ f ( x k , u k ) = ∂ x ( θ k , t k +1 ) ∂ θ k ∂ θ k , ∂ x k ∂ θ k ∂ x k ∂ u k ∂ θ k ∂ u k Implicit function theorem states that ∂ c ∂ θ k + ∂ c ∂ c ∂ θ k + ∂ c = 0 , = 0 − 1 is ∂ θ k ∂ x k ∂ x k ∂ θ k ∂ u k ∂ u k ∂ c Note that ∂ θ k computed in the Newton Hence: − 1 ∂ c − 1 ∂ c iteration, i.e. it comes for ∂ θ k = − ∂ c ∂ θ k = − ∂ c , free !! ∂ x k ∂ θ k ∂ x k ∂ u k ∂ θ k ∂ u k 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 12 / 24

  53. Outline Polynomial interpolation 1 Collocation-based integration 2 Collocation in multiple-shooting 3 Direct Collocation 4 NLP from direct collocation 5 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 13 / 24

  54. Collocation-based integrators in Multiple-shooting Collocation-based integrator solves: c ( x k , u k , θ k ) = 0 on each time interval [ t k , t k +1 ], provides: f ( x k , u k ) = x ( θ k , t k +1 ) with sensitivities. t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 14 / 24

  55. Collocation-based integrators in Multiple-shooting NLP with multiple-shooting Collocation-based integrator solves: min Φ ( w ) c ( x k , u k , θ k ) = 0 w   x 0 − ¯ x 0 on each time interval [ t k , t k +1 ], f ( x 0 , u 0 ) − x 1   s.t. g ( w ) = provides:   ...   f ( x N − 1 , u N − 1 ) − x N f ( x k , u k ) = x ( θ k , t k +1 ) where w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } with sensitivities. t k t k +1 θ k, 3 f ( x k , u k ) 3 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 14 / 24

  56. Collocation-based integrators in Multiple-shooting NLP with multiple-shooting Collocation-based integrator solves: min Φ ( w ) c ( x k , u k , θ k ) = 0 w   x 0 − ¯ x 0 on each time interval [ t k , t k +1 ], f ( x 0 , u 0 ) − x 1   s.t. g ( w ) = provides:   ...   f ( x N − 1 , u N − 1 ) − x N f ( x k , u k ) = x ( θ k , t k +1 ) where w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } with sensitivities. t k t k +1 θ k, 3 f ( x k , u k ) 3 NLP solves: 2 θ k, 2 x k +1 ∇ w L ( w , λ ) = 0 1 x ( θ k , t ) g ( w ) = 0 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 14 / 24

  57. Collocation-based integrators in Multiple-shooting NLP with multiple-shooting Collocation-based integrator solves: min Φ ( w ) c ( x k , u k , θ k ) = 0 w   x 0 − ¯ x 0 on each time interval [ t k , t k +1 ], f ( x 0 , u 0 ) − x 1   s.t. g ( w ) = provides:   ...   f ( x N − 1 , u N − 1 ) − x N f ( x k , u k ) = x ( θ k , t k +1 ) where w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } with sensitivities. t k t k +1 θ k, 3 f ( x k , u k ) 3 NLP solves: 2 θ k, 2 x k +1 ∇ w L ( w , λ ) = 0 1 x ( θ k , t ) g ( w ) = 0 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 Collocation-based integrator inside the NLP becomes a two-level Newton scheme !! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 14 / 24

  58. Collocation-based integrators in Multiple-shooting (cont’) NLP solver ∇ w L ( w , λ ) = 0 g ( w ) = 0 w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } . . . . . . 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 15 / 24

  59. Collocation-based integrators in Multiple-shooting (cont’) NLP level NLP solver Constraints g = 0 ∇ w L ( w , λ ) = 0 Newton iterations g ( w ) = 0 (SQP/IP) w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } . . . . . . 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 15 / 24

  60. Collocation-based integrators in Multiple-shooting (cont’) NLP level NLP solver Constraints g = 0 ∇ w L ( w , λ ) = 0 Newton iterations g ( w ) = 0 (SQP/IP) w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } x 0 , u 0 x k , u k Integrator [ t 0 , t 1 ] Integrator [ t k , t k +1 ] c ( x 0 , u 0 , θ 0 ) = 0 . . . c ( x k , u k , θ k ) = 0 . . . f ( x 0 , u 0 ) = x ( θ 0 , t 1 ) f ( x k , u k ) = x ( θ k , t k +1 ) with sensitivities with sensitivities 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 15 / 24

  61. Collocation-based integrators in Multiple-shooting (cont’) NLP level NLP solver Constraints g = 0 ∇ w L ( w , λ ) = 0 Newton iterations g ( w ) = 0 (SQP/IP) w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } x 0 , u 0 x k , u k Integrator level Integrator [ t 0 , t 1 ] Integrator [ t k , t k +1 ] Constraints c = 0 c ( x 0 , u 0 , θ 0 ) = 0 . . . c ( x k , u k , θ k ) = 0 . . . Newton iterations f ( x 0 , u 0 ) = x ( θ 0 , t 1 ) f ( x k , u k ) = x ( θ k , t k +1 ) with sensitivities with sensitivities 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 15 / 24

  62. Collocation-based integrators in Multiple-shooting (cont’) NLP level NLP solver Constraints g = 0 ∇ w L ( w , λ ) = 0 Newton iterations g ( w ) = 0 (SQP/IP) w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } x 0 , u 0 x k , u k Integrator level Integrator [ t 0 , t 1 ] Integrator [ t k , t k +1 ] Constraints c = 0 c ( x 0 , u 0 , θ 0 ) = 0 . . . c ( x k , u k , θ k ) = 0 . . . Newton iterations f ( x 0 , u 0 ) = x ( θ 0 , t 1 ) f ( x k , u k ) = x ( θ k , t k +1 ) with sensitivities with sensitivities 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 15 / 24

  63. Collocation-based integrators in Multiple-shooting (cont’) NLP level NLP solver Constraints g = 0 ∇ w L ( w , λ ) = 0 Newton iterations g ( w ) = 0 (SQP/IP) w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } x 0 , u 0 x k , u k Integrator level Integrator [ t 0 , t 1 ] Integrator [ t k , t k +1 ] Constraints c = 0 c ( x 0 , u 0 , θ 0 ) = 0 . . . c ( x k , u k , θ k ) = 0 . . . Newton iterations f ( x 0 , u 0 ) = x ( θ 0 , t 1 ) f ( x k , u k ) = x ( θ k , t k +1 ) with sensitivities with sensitivities Constraints are solved at the NLP and at the integrator level separately !! 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 15 / 24

  64. Collocation-based integrators in Multiple-shooting (cont’) NLP level NLP solver Constraints g = 0 ∇ w L ( w , λ ) = 0 Newton iterations g ( w ) = 0 (SQP/IP) w = { x 0 , u 0 , ..., x N − 1 , u N − 1 , x N } x 0 , u 0 x k , u k Integrator level Integrator [ t 0 , t 1 ] Integrator [ t k , t k +1 ] Constraints c = 0 c ( x 0 , u 0 , θ 0 ) = 0 . . . c ( x k , u k , θ k ) = 0 . . . Newton iterations f ( x 0 , u 0 ) = x ( θ 0 , t 1 ) f ( x k , u k ) = x ( θ k , t k +1 ) with sensitivities with sensitivities Constraints are solved at the NLP and at the integrator level separately !! ... what about handling them altogether in the NLP ?!? 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 15 / 24

  65. Outline Polynomial interpolation 1 Collocation-based integration 2 Collocation in multiple-shooting 3 Direct Collocation 4 NLP from direct collocation 5 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 16 / 24

  66. Direct collocation - Give all constraints to the NLP solver On each interval [ t k , t k +1 ] x = F ( x , u k ) ˙ is approximated using: K � x ( θ k , t ) = θ k , i · P k , i ( t ) ���� � �� � i =0 parameters polynomials Note: � � x θ k , i , t k , i = θ k , i K + 1 degrees of freedom per state. t k t k +1 θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) 0 θ k, 1 θ k, 0 -1 x k t k, 0 t k, 1 t k, 2 t k, 3 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  67. Direct collocation - Give all constraints to the NLP solver On each interval [ t k , t k +1 ] x = F ( x , u k ) ˙ is approximated using: K � x ( θ k , t ) = θ k , i · P k , i ( t ) ���� � �� � i =0 parameters polynomials Note: � � x θ k , i , t k , i = θ k , i K + 1 degrees of freedom per state. Integration constraints ( i = 1 , ..., K ) t k t k +1 � = F � x � θ k , t k , i � , u k ∂ ∂ t x � θ k , t k , i � θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) i.e. 0 K θ k, 1 � θ k, 0 P k , j ( t k , i ) = F � θ k , i , u k � θ k , j ˙ -1 x k t k, 0 t k, 1 t k, 2 t k, 3 j =0 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  68. Direct collocation - Give all constraints to the NLP solver NLP with direct collocation On each interval [ t k , t k +1 ] min Φ ( w ) w x = F ( x , u k ) ˙     is approximated using:       K s.t. g ( w ) =   �   x ( θ k , t ) = θ k , i · P k , i ( t )     ���� � �� � i =0   parameters polynomials Note: � � x θ k , i , t k , i = θ k , i K + 1 degrees of freedom per state. Integration constraints ( i = 1 , ..., K ) t k t k +1 � = F � x � θ k , t k , i � , u k ∂ ∂ t x � θ k , t k , i � θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) i.e. 0 K θ k, 1 � θ k, 0 P k , j ( t k , i ) = F � θ k , i , u k � θ k , j ˙ -1 x k t k, 0 t k, 1 t k, 2 t k, 3 j =0 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  69. Direct collocation - Give all constraints to the NLP solver NLP with direct collocation On each interval [ t k , t k +1 ] min Φ ( w ) w x = F ( x , u k ) ˙   θ 0 , 0 − ¯ x 0   is approximated using:       K s.t. g ( w ) =   �   x ( θ k , t ) = θ k , i · P k , i ( t )     ���� � �� � i =0   parameters polynomials Note: � � x θ k , i , t k , i = θ k , i Initial conditions ¯ x 0 K + 1 degrees of freedom per state. Integration constraints ( i = 1 , ..., K ) t k t k +1 � = F � x � θ k , t k , i � , u k ∂ ∂ t x � θ k , t k , i � θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) i.e. 0 K θ k, 1 � θ k, 0 P k , j ( t k , i ) = F � θ k , i , u k � θ k , j ˙ -1 x k t k, 0 t k, 1 t k, 2 t k, 3 j =0 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  70. Direct collocation - Give all constraints to the NLP solver NLP with direct collocation On each interval [ t k , t k +1 ] min Φ ( w ) w x = F ( x , u k ) ˙   θ 0 , 0 − ¯ x 0 x ( θ 0 , t 1 ) − θ 1 , 0   is approximated using:       K s.t. g ( w ) =   �   x ( θ k , t ) = θ k , i · P k , i ( t )     ���� � �� � i =0   parameters polynomials Note: � � x θ k , i , t k , i = θ k , i Continuity constraints ( ≡ shooting gaps) K + 1 degrees of freedom per state. Integration constraints ( i = 1 , ..., K ) t k t k +1 � = F � x � θ k , t k , i � , u k ∂ ∂ t x � θ k , t k , i � θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) i.e. 0 K θ k, 1 � θ k, 0 P k , j ( t k , i ) = F � θ k , i , u k � θ k , j ˙ -1 x k t k, 0 t k, 1 t k, 2 t k, 3 j =0 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  71. Direct collocation - Give all constraints to the NLP solver NLP with direct collocation On each interval [ t k , t k +1 ] min Φ ( w ) w x = F ( x , u k ) ˙   θ 0 , 0 − ¯ x 0 x ( θ 0 , t 1 ) − θ 1 , 0   is approximated using: � �  − � K  j =0 θ 0 , j ˙ F θ 0 , i , u 0 P 0 , j ( t 0 , i )     K s.t. g ( w ) =   �   x ( θ k , t ) = θ k , i · P k , i ( t )     ���� � �� � i =0   parameters polynomials Note: � � x θ k , i , t k , i = θ k , i Integration constraints for k = 0 K + 1 degrees of freedom per state. Integration constraints ( i = 1 , ..., K ) t k t k +1 � = F � x � θ k , t k , i � , u k ∂ ∂ t x � θ k , t k , i � θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) i.e. 0 K θ k, 1 � θ k, 0 P k , j ( t k , i ) = F � θ k , i , u k � θ k , j ˙ -1 x k t k, 0 t k, 1 t k, 2 t k, 3 j =0 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  72. Direct collocation - Give all constraints to the NLP solver NLP with direct collocation On each interval [ t k , t k +1 ] min Φ ( w ) w x = F ( x , u k ) ˙   θ 0 , 0 − ¯ x 0 x ( θ 0 , t 1 ) − θ 1 , 0   is approximated using: � �  − � K  j =0 θ 0 , j ˙ F θ 0 , i , u 0 P 0 , j ( t 0 , i )     K s.t. g ( w ) = ...   �   x ( θ k , t ) = θ k , i · P k , i ( t ) x ( θ k , t k +1 ) − θ k +1 , 0   � − � K   ���� � �� � F � θ k , i , u k j =0 θ k , j ˙ i =0 P k , j ( t k , i )   parameters polynomials ... Note: � � x θ k , i , t k , i = θ k , i Remaining integration constraints k = 1 , ..., N − 1 K + 1 degrees of freedom per state. Integration constraints ( i = 1 , ..., K ) t k t k +1 � = F � x � θ k , t k , i � , u k ∂ ∂ t x � θ k , t k , i � θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) i.e. 0 K θ k, 1 � θ k, 0 P k , j ( t k , i ) = F � θ k , i , u k � θ k , j ˙ -1 x k t k, 0 t k, 1 t k, 2 t k, 3 j =0 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  73. Direct collocation - Give all constraints to the NLP solver NLP with direct collocation On each interval [ t k , t k +1 ] min Φ ( w ) w x = F ( x , u k ) ˙   θ 0 , 0 − ¯ x 0 x ( θ 0 , t 1 ) − θ 1 , 0   is approximated using: � �  − � K  j =0 θ 0 , j ˙ F θ 0 , i , u 0 P 0 , j ( t 0 , i )     K s.t. g ( w ) = ...   �   x ( θ k , t ) = θ k , i · P k , i ( t ) x ( θ k , t k +1 ) − θ k +1 , 0   � − � K   ���� � �� � F � θ k , i , u k j =0 θ k , j ˙ i =0 P k , j ( t k , i )   parameters polynomials ... Note: Decision variables: � � � � x θ k , i , t k , i = θ k , i w = θ 0 , 0 , ..., θ 0 , K , u 0 , ..., θ N − 1 , 0 , ..., θ N − 1 , K , u N − 1 K + 1 degrees of freedom per state. Integration constraints ( i = 1 , ..., K ) t k t k +1 � = F � x � θ k , t k , i � , u k ∂ ∂ t x � θ k , t k , i � θ k, 3 3 f ( x k , u k ) 2 θ k, 2 x k +1 1 x ( θ k , t ) i.e. 0 K θ k, 1 � θ k, 0 P k , j ( t k , i ) = F � θ k , i , u k � θ k , j ˙ -1 x k t k, 0 t k, 1 t k, 2 t k, 3 j =0 -2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 17 / 24

  74. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

  75. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ N = 20 NLP with direct collocation K = 4 with Legendre, order 8 !! N − 1 420 variables � u 2 min 404 constraints k w k =0   θ 0 , 0 − ¯ x 0 Reminder: x ( θ 0 , t 1 ) − θ 1 , 0   � � K  − � K  j =0 θ 0 , j ˙ � F θ 0 , i , u 0 P 0 , j ( t 0 , i )   x ( θ k , t ) = θ k , i · P k , i ( t )   ...   s.t. g ( w ) = = 0   i =0 x ( θ k , t k +1 ) − θ k +1 , 0     � � − � K x ( θ k , t k , i ) = θ k , i j =0 θ k , j ˙  F θ k , i , u k P k , j ( t k , i )     ...  x ( θ N − 1 , t N ) 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

  76. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ Newton step 0 15 4.0 3 3.5 2 10 3.0 1 5 2.5 dx 0 θ 0 dt 2.0 −1 u −5 1.5 −2 1.0 −3 −10 0.5 −4 −15 0.0 −5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 −20 1.5 −25 0.0 0.5 1.0 1.5 2.0 2 1.0 0 dθ x K + 1 = 5 dt −2 0.5 all nodes are initialised −4 0.0 −6 −0.5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

  77. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ Newton step 1 15 4.0 3 3.5 2 10 3.0 1 5 2.5 dx 0 θ 0 dt 2.0 −1 u −5 1.5 −2 1.0 −3 −10 0.5 −4 −15 0.0 −5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 −20 1.5 −25 0.0 0.5 1.0 1.5 2.0 2 1.0 0 dθ x K + 1 = 5 dt −2 0.5 all nodes are initialised −4 0.0 −6 −0.5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

  78. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ Newton step 2 15 4.0 3 3.5 2 10 3.0 1 5 2.5 dx 0 θ 0 dt 2.0 −1 u −5 1.5 −2 1.0 −3 −10 0.5 −4 −15 0.0 −5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 −20 1.5 −25 0.0 0.5 1.0 1.5 2.0 2 1.0 0 dθ x K + 1 = 5 dt −2 0.5 all nodes are initialised −4 0.0 −6 −0.5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

  79. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ Newton step 3 15 4.0 3 3.5 2 10 3.0 1 5 2.5 dx 0 θ 0 dt 2.0 −1 u −5 1.5 −2 1.0 −3 −10 0.5 −4 −15 0.0 −5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 −20 1.5 −25 0.0 0.5 1.0 1.5 2.0 2 1.0 0 dθ x K + 1 = 5 dt −2 0.5 all nodes are initialised −4 0.0 −6 −0.5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

  80. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ Newton step 4 15 4.0 3 3.5 2 10 3.0 1 5 2.5 dx 0 θ 0 dt 2.0 −1 u −5 1.5 −2 1.0 −3 −10 0.5 −4 −15 0.0 −5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 −20 1.5 −25 0.0 0.5 1.0 1.5 2.0 2 1.0 0 dθ x K + 1 = 5 dt −2 0.5 all nodes are initialised −4 0.0 −6 −0.5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

  81. Direct Collocation - Example: swing-up of a pendulum M OCP N − 1 θ � u 2 L min k u 0 ,..., u N − 1 k =0 m s.t. x = F ( x , u k ) , ˙ ∀ t ∈ [ t k , t k +1 ] u � � x x (0) = 0 π 0 0 , x ( t f ) = 0 � � ⊤ ˙ x = x θ x ˙ θ Newton step 5 15 4.0 3 3.5 2 10 3.0 1 5 2.5 dx 0 θ 0 dt 2.0 −1 u −5 1.5 −2 1.0 −3 −10 0.5 −4 −15 0.0 −5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 −20 1.5 −25 0.0 0.5 1.0 1.5 2.0 2 1.0 0 dθ x K + 1 = 5 dt −2 0.5 all nodes are initialised −4 0.0 −6 −0.5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 18 th of February, 2016 S. Gros Optimal Control with DAEs, lecture 8 18 / 24

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Numerical Optimal Control with DAEs Lecture 11: High-Index DAEs S ebastien Gros AWESCO PhD

Numerical Optimal Control with DAEs Lecture 11: High-Index DAEs S ebastien Gros AWESCO PhD

Numerical Optimal Control with DAEs Lecture 11: High-Index DAEs S ebastien Gros AWESCO PhD course 22 nd of February, 2016 S. Gros Optimal Control with DAEs, lecture 11 1 / 25 Objectives of the lecture Why are DAEs not always easyto

2.08k views • 116 slides
Numerical solution of optimal control problems for descriptor systems Volker Mehrmann TU Berlin

Numerical solution of optimal control problems for descriptor systems Volker Mehrmann TU Berlin

Optimal control Applications DAEs OC for ODEs OC for DAEs Riccati Nonlinear problems Numerics Conclusions Numerical solution of optimal control problems for descriptor systems Volker Mehrmann TU Berlin DFG Research Center Institut fr

684 views • 42 slides
Modeling Physics with Differential-Algebraic Equations Lecture 3 Numerical Integration of DAEs

Modeling Physics with Differential-Algebraic Equations Lecture 3 Numerical Integration of DAEs

Modeling Physics with Differential-Algebraic Equations Lecture 3 Numerical Integration of DAEs COMASIC (M2) January 7th, 2019 Khalil Ghorbal k halil.ghorbal@inria.fr K. Ghorbal (INRIA) 1 COMASIC M2 1 / 22 Summary of lecture 2 Index

633 views • 36 slides
Numerical Optimal Control Overview Moritz Diehl Simplified Optimal Control Problem in ODE path

Numerical Optimal Control Overview Moritz Diehl Simplified Optimal Control Problem in ODE path

Numerical Optimal Control Overview Moritz Diehl Simplified Optimal Control Problem in ODE path constraints h ( x , u ) 0 states x ( t ) terminal initial value constraint r ( x ( T )) 0 x 0 r controls u ( t ) 0 t T

865 views • 50 slides
Geometric Numerical Integration of Hamiltonian systems: application to some optimal control

Geometric Numerical Integration of Hamiltonian systems: application to some optimal control

Geometric Numerical Integration of Hamiltonian systems: application to some optimal control problems Philippe Chartier 1 1 IPSO INRIA-Rennes Optimal Control : Algorithms and Applications, May 30-June 1st 2007 logo First examples Hamiltonian

899 views • 62 slides
Numerical Optimal Control Moritz Diehl Optimization in Engineering Center (OPTEC) &amp;

Numerical Optimal Control Moritz Diehl Optimization in Engineering Center (OPTEC) &amp;

Numerical Optimal Control Moritz Diehl Optimization in Engineering Center (OPTEC) &amp; Electrical Engineering Department (ESAT) K.U. Leuven Belgium Simplified Optimal Control Problem in ODE path constraints h ( x, u ) 0 states x ( t

918 views • 53 slides
Geometric and numerical methods in optimal control for the time minimal saturation in Magnetic

Geometric and numerical methods in optimal control for the time minimal saturation in Magnetic

Geometric and numerical methods in optimal control for the time minimal saturation in Magnetic Resonance Imaging DYNAMICS, CONTROL, and GEOMETRY In honor of Bronisaw Jakubczyks 70th birthday 12.09.2018 - 15.09.2018 | Banach Center, Warsaw

324 views • 30 slides
Optimal Control, LQR, Trajectory Optimization Lecture 13 What will you take home today? Intro

Optimal Control, LQR, Trajectory Optimization Lecture 13 What will you take home today? Intro

Optimal Control, LQR, Trajectory Optimization Lecture 13 What will you take home today? Intro Optimal Control Principle of Optimality Bellman Equation Deriving LQR Trajectory Optimization Paper Optimal Control and Reinforcement Learning

486 views • 38 slides
Optimal Control of Hyperbolic Conservation Laws with State Constraints and Convergent Numerical

Optimal Control of Hyperbolic Conservation Laws with State Constraints and Convergent Numerical

Optimal Control of Hyperbolic Conservation Laws with State Constraints and Convergent Numerical Schemes for Adjoints Stefan Ulbrich Department of Mathematics TU Darmstadt Joint work with Paloma Schfer Aguilar, Johann M. Schmitt and Michael

743 views • 70 slides
Optimization-Based Control: Direct Collocation Methods for Trajectory and Policy Optimization CS

Optimization-Based Control: Direct Collocation Methods for Trajectory and Policy Optimization CS

Optimization-Based Control: Direct Collocation Methods for Trajectory and Policy Optimization CS 287: Advanced Robotics, Fall 2019 Guest Lecture Igor Mordatch Overview Previously: Locally optimal control (shooting vs. collocation)

1.54k views • 138 slides
Visual Servoing, Intro Optimal Control Lecture 12 What will you take home today? Visual

Visual Servoing, Intro Optimal Control Lecture 12 What will you take home today? Visual

Visual Servoing, Intro Optimal Control Lecture 12 What will you take home today? Visual Servoing Interaction Matrix Control Law Case-Study: Learning-based approach Introduction Optimal Control Principle of Optimality Bellman Equation

758 views • 47 slides
Numerical approximation for optimal control problems via MPC and HJB Giulia Fabrini Konstanz

Numerical approximation for optimal control problems via MPC and HJB Giulia Fabrini Konstanz

Numerical approximation for optimal control problems via MPC and HJB Giulia Fabrini Konstanz Women In Mathematics 15 May, 2018 G. Fabrini (University of Konstanz) Numerical approximation for OCP 1 / 33 Outline Outline Introduction and

729 views • 33 slides
Interoperating direct and indirect optimal control solvers Olivier Cots, Joint work with J.-B.

Interoperating direct and indirect optimal control solvers Olivier Cots, Joint work with J.-B.

Interoperating direct and indirect optimal control solvers Olivier Cots, Joint work with J.-B. Caillau, P. Martinon and the invaluable help of Inria Sophia SED 9th International Congress on Industrial and Applied Mathematics (ICIAM 2019), July

631 views • 27 slides
Optimal control of the atmospheric re-entry of a space shuttle Emmanuel Trlat 1 1 MAPMO

Optimal control of the atmospheric re-entry of a space shuttle Emmanuel Trlat 1 1 MAPMO

Model Numerical methods System analysis Numerical results Open problems Optimal control of the atmospheric re-entry of a space shuttle Emmanuel Trlat 1 1 MAPMO Universit dOrlans EDF-CEA-INRIA School, 2007 E. Trlat Atmospheric

623 views • 27 slides
Draft EE 8235: Lecture 27 1 Lecture 27: Optimal control of undirected graphs

Draft EE 8235: Lecture 27 1 Lecture 27: Optimal control of undirected graphs

Draft EE 8235: Lecture 27 1 Lecture 27: Optimal control of undirected graphs Single-integrator dynamics x i = u i + d i Relative information exchange with neighbors u i ( t ) = k ij x i ( t ) x j ( t ) j N i

519 views • 15 slides
Draft EE 8235: Lecture 23 1 Lecture 23: Optimal control of distributed systems Linear

Draft EE 8235: Lecture 23 1 Lecture 23: Optimal control of distributed systems Linear

Draft EE 8235: Lecture 23 1 Lecture 23: Optimal control of distributed systems Linear Quadratic Regulator (LQR) Linear: plant Quadratic: performance index Infinite horizon problem Algebraic Riccati Equation (ARE) Spatially

324 views • 11 slides
Green Ammonia pilot plants and their implications for the fertiliser industry IFS webinar, April

Green Ammonia pilot plants and their implications for the fertiliser industry IFS webinar, April

22/04/2020 The forces driving industrial scale-up of Green Ammonia pilot plants and their implications for the fertiliser industry IFS webinar, April 15, 2020 Trevor Brown, Ammonia Energy Association 1 Ammonia Energy Association A global

316 views • 9 slides
Sirius Web 100% open source cloud modeling platform Our Crew Mlanie Bats Stphane

Sirius Web 100% open source cloud modeling platform Our Crew Mlanie Bats Stphane

Sirius Web 100% open source cloud modeling platform Our Crew Mlanie Bats Stphane Bgaudeau CTO @ Obeo Software Architect @ Obeo melanie.bats@obeo.fr stephane.begaudeau@obeo.fr @melaniebats @sbegaudeau Sirius-II Mission Countdown

701 views • 22 slides
CSE 127 Computer Security Deian Stefan, Winter 2019 Lecture 1: Introduction Course Information

CSE 127 Computer Security Deian Stefan, Winter 2019 Lecture 1: Introduction Course Information

CSE 127 Computer Security Deian Stefan, Winter 2019 Lecture 1: Introduction Course Information Lecturer: Deian Stefan Lectures: Mon &amp; Wed 6:30-7:50pm, Solis 104 Discussion: Tue 8:00-8:50pm, Center 105 Office Hours: Mon 1-2pm

881 views • 29 slides
Coulomb Systems: Thermodynamics, Fluctuations, Large Deviations and Rigidity Joel L. Lebowitz

Coulomb Systems: Thermodynamics, Fluctuations, Large Deviations and Rigidity Joel L. Lebowitz

Coulomb Systems: Thermodynamics, Fluctuations, Large Deviations and Rigidity Joel L. Lebowitz Rutgers University Paris, November 2015 Dedicated to the memory of Janco. Friend, colleague and teacher 1 / 36 The properties of macroscopic matter

594 views • 44 slides
Integrating Sustainability Into DCs E-Procurement System By Jonathan Rifkin Sustainable

Integrating Sustainability Into DCs E-Procurement System By Jonathan Rifkin Sustainable

Integrating Sustainability Into DCs E-Procurement System By Jonathan Rifkin Sustainable Purchasing Coordinator, DC Office of Contracting and Procurement May 21, 2014 What happens if I win the lotto I get hit by a bus Building the

557 views • 13 slides
Software-Defined Networks Jennifer Rexford Princeton University Traditional Networks 2 control

Software-Defined Networks Jennifer Rexford Princeton University Traditional Networks 2 control

Software-Defined Networks Jennifer Rexford Princeton University Traditional Networks 2 control plane: distributed algorithms data plane: packet processing decouple control and data planes Software Defined Networks 3 decouple control and

370 views • 20 slides
Run Kubernetes on OpenStack and Bare Metal Fast Ramon Acedo Rodriguez 1 Senior Principal

Run Kubernetes on OpenStack and Bare Metal Fast Ramon Acedo Rodriguez 1 Senior Principal

OPEN INFRASTRUCTURE SUMMIT | SHANGHAI, NOVEMBER 4-6 2019 Run Kubernetes on OpenStack and Bare Metal Fast Ramon Acedo Rodriguez 1 Senior Principal Product Manager, Red Hat Open Hybrid Cloud Vision OPTIONAL SECTION MARKER OR TITLE 2 The

1.02k views • 53 slides
Object-oriented design : principles J.Serrat 102759 Software Design

Object-oriented design : principles J.Serrat 102759 Software Design

Object-oriented design : principles J.Serrat 102759 Software Design http://www.cvc.uab.es/shared/teach/a21291/web/ August 6, 2014 Index Information hiding 1 Dont talk to strangers 2 DRY: Dont repeat yourself 3 SRP: Single

507 views • 30 slides

More recommend