第一部分:残差函数
约定符号:
IMU的真实值:
ω
b
,
a
b
\omega^b,a^b
ωb,ab
IMU的测量值:
ω
~
b
,
a
~
b
\tilde{\omega}^b,\tilde{a}^b
ω~b,a~b
某个时刻
t
t
t,二者的关系:
ω
~
t
b
t
=
ω
t
b
t
+
b
t
g
+
n
t
g
a
~
t
b
t
=
a
t
b
t
+
b
t
a
+
n
t
a
=
q
b
t
w
(
a
t
w
+
g
w
)
+
b
t
a
+
n
t
a
\begin{aligned} \tilde{\omega}^{b_t}_t &=\omega^{b_t}_t+b^g_t+n^g_t \\ \tilde{a}^{b_t}_t &=a^{b_t}_t+b^a_t+n^a_t \\ &= q_{b_tw}(a^w_t+g^w)+b^a_t+n^a_t \end{aligned}
ω~tbta~tbt=ωtbt+btg+ntg=atbt+bta+nta=qbtw(atw+gw)+bta+nta
ps. IMU机体系即
b
b
b系,
w
w
w系表示世界坐标系。
不考虑高斯白噪声项,有:
ω
t
b
t
≈
ω
~
t
b
t
−
b
t
g
a
t
w
≈
q
w
b
t
(
a
~
t
b
t
−
b
t
a
)
−
g
w
a
t
b
t
≈
a
~
t
b
t
−
b
t
a
\begin{aligned} \omega^{b_t}_t &\approx \tilde{\omega}^{b_t}_t -b^g_t \\ a^w_t &\approx q_{wb_t} (\tilde{a}^{b_t}_t - b^a_t)- g^w\\ a^{b_t}_t &\approx \tilde{a}^{b_t}_t - b^a_t \end{aligned}
ωtbtatwatbt≈ω~tbt−btg≈qwbt(a~tbt−bta)−gw≈a~tbt−bta
考虑高斯白噪声项,有:
ω
t
b
t
=
ω
~
t
b
t
−
b
t
g
−
n
t
g
a
t
w
=
q
w
b
t
(
a
~
t
b
t
−
b
t
a
−
n
t
a
)
−
g
w
a
t
b
t
=
a
~
t
b
t
−
b
t
a
−
n
t
a
\begin{aligned} \omega^{b_t}_t &= \tilde{\omega}^{b_t}_t -b^g_t -n^g_t\\ a^w_t &= q_{wb_t} (\tilde{a}^{b_t}_t - b^a_t-n^a_t)- g^w\\ a^{b_t}_t &= \tilde{a}^{b_t}_t - b^a_t-n^a_t \end{aligned}
ωtbtatwatbt=ω~tbt−btg−ntg=qwbt(a~tbt−bta−nta)−gw=a~tbt−bta−nta
【注意】:此处考虑与不考虑高斯噪声,记为了同一个符号
P(ose),V(elocity),Q(uaternion)某一时刻
t
t
t的导数为:
p
˙
w
b
t
=
v
t
w
v
˙
t
w
=
a
t
w
q
˙
w
b
t
=
q
w
b
t
⊗
[
0
1
2
ω
t
b
t
]
\begin{aligned} \dot{p}_{w b_t} &=v^w_t \\ \dot{v}^w_t &=a^w_t \\ \dot{q}_{w b_t}&= {q}_{w b_t}\otimes \begin{bmatrix} 0 \\ \frac{1}{2}\omega^{b_t}_t \end{bmatrix} \end{aligned}
p˙wbtv˙twq˙wbt=vtw=atw=qwbt⊗[021ωtbt]
那么,已知第
i
i
i帧的姿态、位置(相对于世界坐标系),根据IMU的输出,可以递推求得第
j
j
j帧的姿态、位置(相对于世界坐标系):
p
w
b
j
=
p
w
b
i
+
v
i
w
Δ
t
+
∫
∫
t
∈
[
i
,
j
]
a
t
w
d
t
2
v
j
w
=
v
i
w
+
∫
t
∈
[
i
,
j
]
a
t
w
d
t
q
w
b
j
=
∫
t
∈
[
i
,
j
]
q
w
b
t
⊗
[
0
1
2
ω
t
b
t
]
d
t
\begin{aligned} {p}_{w b_j} &= {p}_{w b_i}+v^w_i\Delta t +\int \int_{t \in[i,j]}{a^w_t}dt^2 \\ {v}^w_j &=v^w_i+ \int_{t \in[i,j]}{a^w_t}dt \\ {q}_{w b_j}&= \int_{t \in [i,j]} {q}_{w b_t}\otimes \begin{bmatrix} 0 \\ \frac{1}{2}\omega^{b_t}_t \end{bmatrix}dt \end{aligned}
pwbjvjwqwbj=pwbi+viwΔt+∫∫t∈[i,j]atwdt2=viw+∫t∈[i,j]atwdt=∫t∈[i,j]qwbt⊗[021ωtbt]dt
【问题】姿态的积分?
另一种求得第
j
j
j帧的姿态、位置:
应用
q
w
b
t
=
q
w
b
i
⊗
q
b
i
b
t
{q}_{w b_t}= {q}_{w b_i}\otimes {q}_{b_i b_t}
qwbt=qwbi⊗qbibt,转而去求相对于第
i
i
i帧,第
j
j
j帧的姿态、位置变化。
p
w
b
j
=
p
w
b
i
+
v
i
w
Δ
t
−
1
2
g
w
Δ
t
2
+
q
w
b
i
∫
∫
t
∈
[
i
,
j
]
(
q
b
i
b
t
a
t
b
t
)
d
t
2
v
j
w
=
v
i
w
−
g
w
Δ
t
+
q
w
b
i
∫
t
∈
[
i
,
j
]
(
q
b
i
b
t
a
t
b
t
)
d
t
q
w
b
j
=
q
w
b
i
⊗
∫
t
∈
[
i
,
j
]
q
b
i
b
t
⊗
[
0
1
2
ω
t
b
t
]
d
t
\begin{aligned} {p}_{w b_j} &= {p}_{w b_i}+v^w_i\Delta t -\frac{1}{2}g^w\Delta t^2 + {q}_{w b_i} \int \int_{t \in[i,j]} ({q}_{b_i b_t}{a^{b_t}_t}) dt^2 \\ {v}^w_j &=v^w_i-g^w\Delta t +{q}_{w b_i} \int_{t \in[i,j]} ({q}_{b_i b_t}{a^{b_t}_t}) dt \\ {q}_{w b_j}&={q}_{w b_i}\otimes \int_{t \in [i,j]} {q}_{b_i b_t}\otimes \begin{bmatrix} 0 \\ \frac{1}{2}\omega^{b_t}_t \end{bmatrix}dt \end{aligned}
pwbjvjwqwbj=pwbi+viwΔt−21gwΔt2+qwbi∫∫t∈[i,j](qbibtatbt)dt2=viw−gwΔt+qwbi∫t∈[i,j](qbibtatbt)dt=qwbi⊗∫t∈[i,j]qbibt⊗[021ωtbt]dt
【注】四元数之间的乘、四元数与矢量之间的乘
第二种积分的好处在于,即使优化过程使得第
i
i
i帧的姿态、位置发生变化,即
q
w
b
i
q_{wb_i}
qwbi和
p
w
b
i
p_{wb_i}
pwbi变化,但积分部分并不变化!!
IMU预积分以第
i
i
i帧为参考,将一段时间
t
∈
[
i
,
j
]
t\in [i,j]
t∈[i,j]的IMU数据积分起来,为了简化表达,定义三个符号(表示预积分量):
α
b
i
b
j
=
∫
∫
t
∈
[
i
,
j
]
(
q
b
i
b
t
a
t
b
t
)
d
t
2
β
b
i
b
j
=
∫
t
∈
[
i
,
j
]
(
q
b
i
b
t
a
t
b
t
)
d
t
q
b
i
b
j
=
∫
t
∈
[
i
,
j
]
q
b
i
b
t
⊗
[
0
1
2
ω
t
b
t
]
d
t
\begin{aligned} \alpha_{b_ib_j} &= \int \int_{t \in[i,j]} ({q}_{b_i b_t}{a^{b_t}_t}) dt^2 \\ \beta_{b_ib_j} &= \int_{t \in[i,j]} ({q}_{b_i b_t}{a^{b_t}_t}) dt \\ q_{b_ib_j} &= \int_{t \in [i,j]} {q}_{b_i b_t}\otimes \begin{bmatrix} 0 \\ \frac{1}{2}\omega^{b_t}_t \end{bmatrix}dt \end{aligned}
αbibjβbibjqbibj=∫∫t∈[i,j](qbibtatbt)dt2=∫t∈[i,j](qbibtatbt)dt=∫t∈[i,j]qbibt⊗[021ωtbt]dt
【注】在纯惯导里,第三项被称为姿态变化四元数。
综上,整理一下(用新符号表示第
j
j
j帧的姿态、位置):
[
p
w
b
j
v
j
w
q
w
b
j
b
j
a
b
j
g
]
=
[
p
w
b
i
+
v
i
w
Δ
t
−
1
2
g
w
Δ
t
2
+
q
w
b
i
α
b
i
b
j
v
i
w
−
g
w
Δ
t
+
q
w
b
i
β
b
i
b
j
q
w
b
i
⊗
q
b
i
b
j
b
i
a
b
i
g
]
\begin{aligned} \begin{bmatrix} p_{wb_j} \\ v^w_j \\ q_{wb_j} \\ b^a_j \\ b^g_j \end{bmatrix} = \begin{bmatrix} {p}_{w b_i}+v^w_i\Delta t -\frac{1}{2}g^w\Delta t^2 + {q}_{w b_i} \alpha_{b_ib_j} \\ v^w_i-g^w\Delta t +{q}_{w b_i} \beta_{b_ib_j}\\ {q}_{w b_i}\otimes q_{b_ib_j} \\ b^a_i \\ b^g_i \end{bmatrix} \end{aligned}
⎣⎢⎢⎢⎢⎡pwbjvjwqwbjbjabjg⎦⎥⎥⎥⎥⎤=⎣⎢⎢⎢⎢⎡pwbi+viwΔt−21gwΔt2+qwbiαbibjviw−gwΔt+qwbiβbibjqwbi⊗qbibjbiabig⎦⎥⎥⎥⎥⎤
【注意】预积分量,既可以预测(即以递推形式得到)【记为】
α
^
b
i
b
j
,
β
^
b
i
b
j
,
q
^
b
i
b
j
,
\hat{\alpha}_{b_ib_j}, \hat{\beta}_{b_ib_j}, \hat{q}_{b_ib_j},
α^bibj,β^bibj,q^bibj,,又可以被观测(即通过IMU数据的积分得到)【仍记为】
α
b
i
b
j
,
β
b
i
b
j
,
q
b
i
b
j
,
{\alpha}_{b_ib_j}, {\beta}_{b_ib_j}, {q}_{b_ib_j},
αbibj,βbibj,qbibj,,因此使用二者的差值构造IMU的残差函数:
r
b
=
[
r
p
r
v
r
q
r
b
a
r
b
g
]
=
[
α
^
b
i
b
j
−
α
b
i
b
j
β
^
b
i
b
j
−
β
b
i
b
j
1
2
[
q
^
b
j
b
i
⊗
q
b
i
b
j
]
x
y
z
b
j
a
−
b
i
a
b
j
g
−
b
i
g
]
\begin{aligned} r_b= \begin{bmatrix} r_p \\ r_v \\ r_q \\ r_{b_a} \\ r_{b_g} \end{bmatrix} = \begin{bmatrix} \hat{\alpha}_{b_ib_j} -{\alpha}_{b_ib_j} \\ \hat{\beta}_{b_ib_j} - {\beta}_{b_ib_j}\\ \frac{1}{2}[\hat{q}_{b_jb_i} \otimes {q}_{b_ib_j}]_{xyz} \\ b^a_j - b^a_i \\ b^g_j - b^g_i \end{bmatrix} \end{aligned}
rb=⎣⎢⎢⎢⎢⎡rprvrqrbarbg⎦⎥⎥⎥⎥⎤=⎣⎢⎢⎢⎢⎡α^bibj−αbibjβ^bibj−βbibj21[q^bjbi⊗qbibj]xyzbja−biabjg−big⎦⎥⎥⎥⎥⎤
其中,
[
α
^
b
i
b
j
β
^
b
i
b
j
q
^
b
j
b
i
]
=
[
q
b
i
w
(
p
w
b
j
−
p
w
b
i
−
v
i
w
Δ
t
+
1
2
g
w
Δ
t
2
)
q
b
i
w
(
v
j
w
−
v
i
w
+
g
w
Δ
t
)
q
b
j
w
⊗
q
w
b
i
]
\begin{aligned} \begin{bmatrix} \hat \alpha_{b_ib_j} \\ \hat \beta_{b_ib_j} \\ \hat q_{b_jb_i} \end{bmatrix} = \begin{bmatrix} {q}_{ b_i w} (p_{wb_j}- {p}_{w b_i} - v^w_i\Delta t +\frac{1}{2}g^w\Delta t^2) \\ {q}_{b_iw} (v^w_j - v^w_i + g^w\Delta t )\\ {q}_{ b_j w}\otimes q_{wb_i} \end{bmatrix} \end{aligned}
⎣⎡α^bibjβ^bibjq^bjbi⎦⎤=⎣⎡qbiw(pwbj−pwbi−viwΔt+21gwΔt2)qbiw(vjw−viw+gwΔt)qbjw⊗qwbi⎦⎤
一个IMU量测数据的噪声强度通过Allan方差能够标定,现在我们更关心:目前我们使用的预积分量(一段时间内,即第
i
i
i帧到第
j
j
j帧多个IMU数据的积分结果)的方差。
【补充性质】:已知两个随机变量之间的线性关系
y
=
A
x
y=Ax
y=Ax,若有
x
∈
N
(
0
,
Σ
x
)
x\in N(0,\Sigma_x)
x∈N(0,Σx),则可推导得
y
∈
N
(
0
,
Σ
y
)
y\in N(0,\Sigma_y)
y∈N(0,Σy)。
μ
y
=
E
(
A
x
)
=
A
μ
x
=
0
\mu_y=E(Ax)=A\mu_x=0
μy=E(Ax)=Aμx=0
Σ
y
=
E
(
(
A
x
)
(
A
x
)
T
)
=
E
(
A
x
x
T
A
T
)
=
A
Σ
x
A
T
\begin{aligned} \Sigma_y &=E((Ax)(Ax)^T)\\ &= E(Axx^TA^T)\\ &= A\Sigma_xA^T \end{aligned}
Σy=E((Ax)(Ax)T)=E(AxxTAT)=AΣxAT
如果,我们回答上述问题,需要明确任一IMU采样时刻
k
k
k到下一时刻
k
+
1
k+1
k+1之间的随机变量,以及线性关系:
时刻
k
+
1
k+1
k+1的随机变量记为:
η
k
+
1
=
[
δ
α
b
k
+
1
b
k
+
1
′
,
δ
θ
b
k
+
1
b
k
+
1
′
,
δ
β
b
k
+
1
b
k
+
1
′
,
δ
b
b
k
+
1
a
,
δ
b
b
k
+
1
g
]
T
\eta_{k+1}=[\delta \alpha_{b_{k+1}b'_{k+1}},\delta \theta_ {b_{k+1}b'_{k+1}},\delta \beta_{b_{k+1}b'_{k+1}},\delta b^a_{b_{k+1}},\delta b^g_{b_{k+1}}]^T
ηk+1=[δαbk+1bk+1′,δθbk+1bk+1′,δβbk+1bk+1′,δbbk+1a,δbbk+1g]T,
时刻
k
k
k的随机变量记为:
η
k
=
[
δ
α
b
k
b
k
′
,
δ
θ
b
k
b
k
′
,
δ
β
b
k
b
k
′
,
δ
b
b
k
a
,
δ
b
b
k
g
]
T
\eta_{k}=[\delta \alpha_{b_{k}b'_{k}},\delta \theta_ {b_{k}b'_{k}},\delta \beta_{b_{k}b'_{k}},\delta b^a_{b_{k}},\delta b^g_{b_{k}}]^T
ηk=[δαbkbk′,δθbkbk′,δβbkbk′,δbbka,δbbkg]T,
噪声项:
n
k
=
[
n
k
a
,
n
k
g
,
n
k
+
1
a
,
n
k
+
1
g
,
n
b
k
a
,
n
b
k
g
]
T
n_{k}=[n^a_k,n^g_k,n^a_{k+1},n^g_{k+1},n_{b^a_k},n_{b^g_k}]^T
nk=[nka,nkg,nk+1a,nk+1g,nbka,nbkg]T
三者之间的关系:
η
k
+
1
=
F
η
k
+
G
n
k
\eta_{k+1}=F\eta_{k}+Gn_k
ηk+1=Fηk+Gnk
若
η
k
∈
N
(
0
,
Σ
η
k
)
\eta_k\in N(0,\Sigma_{\eta_k})
ηk∈N(0,Σηk),
n
k
∈
N
(
0
,
Σ
n
k
)
n_k\in N(0,\Sigma_{n_k})
nk∈N(0,Σnk),则
η
k
+
1
∈
N
(
0
,
Σ
η
k
+
1
)
\eta_{k+1}\in N(0,\Sigma_{\eta_{k+1}})
ηk+1∈N(0,Σηk+1)。其中,
Σ
η
k
+
1
=
F
Σ
η
k
F
T
+
G
Σ
n
k
G
T
\Sigma_{\eta_{k+1}}=F\Sigma_{\eta_{k}}F^T+G\Sigma_{n_{k}}G^T
Σηk+1=FΣηkFT+GΣnkGT
待求取的是
δ
α
b
k
+
1
b
k
+
1
′
\delta \alpha_{b_{k+1}b'_{k+1}}
δαbk+1bk+1′等和
δ
α
b
k
b
k
′
\delta \alpha_{b_{k}b'_{k}}
δαbkbk′等之间的关系,
目前根据预积分连续形式得到的离散形式可以反应:
α
b
i
b
k
+
1
′
\alpha_{b_{i}b'_{k+1}}
αbibk+1′等和
α
b
i
b
k
′
\alpha_{b_{i}b'_{k}}
αbibk′之间的关系,即:
已知,
α
b
i
b
j
=
∫
∫
t
∈
[
i
,
j
]
(
q
b
i
b
t
a
t
b
t
)
d
t
2
β
b
i
b
j
=
∫
t
∈
[
i
,
j
]
(
q
b
i
b
t
a
t
b
t
)
d
t
q
b
i
b
j
=
∫
t
∈
[
i
,
j
]
q
b
i
b
t
⊗
[
0
1
2
ω
t
b
t
]
d
t
\begin{aligned} \alpha_{b_ib_j} &= \int \int_{t \in[i,j]} ({q}_{b_i b_t}{a^{b_t}_t}) dt^2 \\ \beta_{b_ib_j} &= \int_{t \in[i,j]} ({q}_{b_i b_t}{a^{b_t}_t}) dt \\ q_{b_ib_j} &= \int_{t \in [i,j]} {q}_{b_i b_t}\otimes \begin{bmatrix} 0 \\ \frac{1}{2}\omega^{b_t}_t \end{bmatrix}dt \end{aligned}
αbibjβbibjqbibj=∫∫t∈[i,j](qbibtatbt)dt2=∫t∈[i,j](qbibtatbt)dt=∫t∈[i,j]qbibt⊗[021ωtbt]dt
则:
α
b
i
b
k
+
1
=
α
b
i
b
k
+
β
b
i
b
k
δ
t
+
1
2
a
δ
t
2
β
b
i
b
k
+
1
=
β
b
i
b
k
+
a
δ
t
q
b
i
b
k
+
1
=
q
b
i
b
k
⊗
[
0
1
2
ω
δ
t
]
\begin{aligned} \alpha_{b_ib_{k+1}} &= \alpha_{b_ib_{k}} + \beta_{b_ib_k}\delta t +\frac{1}{2}a\delta t^2\\ \beta_{b_ib_{k+1}} &= \beta_{b_ib_{k}} +a\delta t \\ q_{b_ib_{k+1}} &= q_{b_ib_{k}}\otimes \begin{bmatrix} 0 \\ \frac{1}{2}\omega \delta t \end{bmatrix} \end{aligned}
αbibk+1βbibk+1qbibk+1=αbibk+βbibkδt+21aδt2=βbibk+aδt=qbibk⊗[021ωδt]
使用中值积分法(考虑高斯噪声):
ω
=
1
2
(
ω
k
b
k
+
ω
k
+
1
b
k
+
1
)
=
1
2
[
(
ω
~
k
b
k
−
b
k
g
−
n
k
g
)
+
(
ω
~
k
+
1
b
k
+
1
−
b
k
+
1
g
−
n
k
+
1
g
)
]
a
=
1
2
(
a
k
b
k
+
a
k
+
1
b
k
+
1
)
=
1
2
[
(
q
w
b
k
(
a
~
k
b
k
−
b
k
a
−
n
k
a
)
−
g
w
)
+
(
q
w
b
k
+
1
(
a
~
k
+
1
b
k
+
1
−
b
k
+
1
a
−
n
k
+
1
a
)
−
g
w
)
]
\begin{aligned} \omega &= \frac{1}{2}(\omega^{b_k}_k+\omega^{b_{k+1}}_{k+1}) \\ &= \frac{1}{2}[(\tilde{\omega}^{b_k}_k -b^g_k -n^g_k)+(\tilde{\omega}^{b_{k+1}}_{k+1} -b^g_{k+1} -n^g_{k+1})]\\ a &= \frac{1}{2}(a^{b_k}_k+a^{b_{k+1}}_{k+1}) \\ &=\frac{1}{2}[(q_{wb_k} (\tilde{a}^{b_k}_k - b^a_k-n^a_k)- g^w)+(q_{wb_{k+1}} (\tilde{a}^{b_{k+1}}_{k+1} - b^a_{k+1}-n^a_{k+1})- g^w)] \end{aligned}
ωa=21(ωkbk+ωk+1bk+1)=21[(ω~kbk−bkg−nkg)+(ω~k+1bk+1−bk+1g−nk+1g)]=21(akbk+ak+1bk+1)=21[(qwbk(a~kbk−bka−nka)−gw)+(qwbk+1(a~k+1bk+1−bk+1a−nk+1a)−gw)]
注意此处,1.噪声项可加可减,使用加;2.预积分时
g
w
g^w
gw被提到了积分之外;3.预积分相对于
b
i
b_i
bi系,故可写为:
ω
=
1
2
[
(
ω
~
k
b
k
−
b
k
g
+
n
k
g
)
+
(
ω
~
k
+
1
b
k
+
1
−
b
k
+
1
g
+
n
k
+
1
g
)
]
a
=
1
2
[
q
b
i
b
k
(
a
~
k
b
k
−
b
k
a
+
n
k
a
)
+
q
b
i
b
k
+
1
(
a
~
k
+
1
b
k
+
1
−
b
k
+
1
a
+
n
k
+
1
a
)
]
\begin{aligned} \omega &= \frac{1}{2}[(\tilde{\omega}^{b_k}_k -b^g_k +n^g_k)+(\tilde{\omega}^{b_{k+1}}_{k+1}-b^g_{k+1} +n^g_{k+1})]\\ a &=\frac{1}{2}[q_{b_ib_k} (\tilde{a}^{b_k}_k - b^a_k+n^a_k)+q_{b_ib_{k+1}} (\tilde{a}^{b_{k+1}}_{k+1}- b^a_{k+1}+n^a_{k+1})] \end{aligned}
ωa=21[(ω~kbk−bkg+nkg)+(ω~k+1bk+1−bk+1g+nk+1g)]=21[qbibk(a~kbk−bka+nka)+qbibk+1(a~k+1bk+1−bk+1a+nk+1a)]
其中,
b
k
+
1
a
=
b
k
a
+
n
b
k
a
δ
t
b
k
+
1
g
=
b
k
g
+
n
b
k
g
δ
t
b^a_{k+1}=b^a_k+n_{b^a_k}\delta t \\ b^g_{k+1}=b^g_k+n_{b^g_k}\delta t
bk+1a=bka+nbkaδtbk+1g=bkg+nbkgδt
至此,和上一节视觉雅可比公式的推导类似了:即求
F
=
∂
[
α
b
i
b
k
+
1
′
,
θ
b
i
b
k
+
1
′
,
β
b
i
b
k
+
1
′
,
b
b
k
+
1
a
,
b
b
k
+
1
g
]
T
∂
[
δ
α
b
k
b
k
′
,
δ
θ
b
k
b
k
′
,
δ
β
b
k
b
k
′
,
δ
b
b
k
a
,
δ
b
b
k
g
]
T
F=\frac{ \partial [\alpha_{b_{i}b'_{k+1}},\theta_ {b_{i}b'_{k+1}},\beta_{b_{i}b'_{k+1}},b^a_{b_{k+1}},b^g_{b_{k+1}}]^T} { \partial [\delta \alpha_{b_{k}b'_{k}},\delta \theta_ {b_{k}b'_{k}},\delta \beta_{b_{k}b'_{k}},\delta b^a_{b_{k}},\delta b^g_{b_{k}}]^T}
F=∂[δαbkbk′,δθbkbk′,δβbkbk′,δbbka,δbbkg]T∂[αbibk+1′,θbibk+1′,βbibk+1′,bbk+1a,bbk+1g]T
及
G
=
∂
[
α
b
i
b
k
+
1
′
,
θ
b
i
b
k
+
1
′
,
β
b
i
b
k
+
1
′
,
b
b
k
+
1
a
,
b
b
k
+
1
g
]
T
∂
[
n
k
a
,
n
k
g
,
n
k
+
1
a
,
n
k
+
1
g
,
n
b
k
a
,
n
b
k
g
]
T
G=\frac{ \partial [\alpha_{b_{i}b'_{k+1}},\theta_ {b_{i}b'_{k+1}},\beta_{b_{i}b'_{k+1}},b^a_{b_{k+1}},b^g_{b_{k+1}}]^T} { \partial [n^a_k,n^g_k,n^a_{k+1},n^g_{k+1},n_{b^a_k},n_{b^g_k}]^T}
G=∂[nka,nkg,nk+1a,nk+1g,nbka,nbkg]T∂[αbibk+1′,θbibk+1′,βbibk+1′,bbk+1a,bbk+1g]T
可以看出来,这里共有
5
×
5
×
2
=
50
5\times 5 \times 2=50
5×5×2=50个小块,故以F.1.1 ~ F.1.5 ~ … F.5.5 ~ … G.5.5顺序依次求解。
残差函数已经得到,下面明确需要对哪些状态求雅可比?