diff --git a/mr/ub8/localize_jp/localize.m b/mr/ub8/localize_jp/localize.m
index 1293290..b9c16a1 100644
--- a/mr/ub8/localize_jp/localize.m
+++ b/mr/ub8/localize_jp/localize.m
@@ -8,7 +8,7 @@
 % System noise:
 R = diag([0.02^2, 0.02^2, 0.02^2]);
 % Measurement noise (1D measurements):
-Q = 0.5^2;
+Q = 0.002^2;
 % Construct vector with control input:
 uhc = repmat([0.1; 0.105], 1, 63);
 ust = repmat([0.1; 0.1], 1, 30);
@@ -67,7 +67,7 @@
     [z_exp, H] = measurement(curr_mean);
    
     % Kalman gain:
-    K = curr_cov * (H') * inv(H*curr_cov*(H') + Q);
+    K = curr_cov * (H') * inv(H*curr_cov*(H') + Q)
     curr_mean = curr_mean + K*(z_meas(:,i) - z_exp);
     curr_cov = (eye(3) - K*H)*curr_cov;
     
diff --git a/mr/ub8/localize_jp/motion_diff.m b/mr/ub8/localize_jp/motion_diff.m
index 848f892..e22de18 100644
--- a/mr/ub8/localize_jp/motion_diff.m
+++ b/mr/ub8/localize_jp/motion_diff.m
@@ -16,8 +16,12 @@
 else
     r=(sl+sr)/(sr-sl)*l/2;
     dtheta=(sr-sl)/l;
-    xnew=x+[r*(sin(theta+dtheta)-sin(theta));r*(-cos(theta+dtheta)+cos(theta));dtheta];
-    G=jacobian([x1;x2;x3]+[r*(sin(x3+dtheta)-sin(x3));r*(-cos(x3+dtheta)+cos(x3));dtheta],[x1;x2;x3]);
+    xnew=x+[r*(sin(theta+dtheta)-sin(theta));
+        r*(-cos(theta+dtheta)+cos(theta));
+        dtheta];
+    G=jacobian([x1;x2;x3]+[r*(sin(x3+dtheta)-sin(x3));
+        r*(-cos(x3+dtheta)+cos(x3));
+        dtheta],[x1;x2;x3]);
 end
 x3=theta;
 G=eval(G);
diff --git a/mr/ub8/mr8.pdf b/mr/ub8/mr8.pdf
index 4a38336..091b875 100644
--- a/mr/ub8/mr8.pdf
+++ b/mr/ub8/mr8.pdf
Binary files differ
diff --git a/mr/ub8/mr8.tex b/mr/ub8/mr8.tex
index ff50405..0b37338 100644
--- a/mr/ub8/mr8.tex
+++ b/mr/ub8/mr8.tex
@@ -14,6 +14,8 @@
 \usepackage{tikz} %TikZ ist kein Zeichenprogramm
 \usetikzlibrary{trees,automata,arrows,shapes}
 \usepackage{qtree}
+\usepackage{listings}
+\lstset{language=Matlab}
 
 \pagestyle{empty}
 
@@ -72,38 +74,62 @@
 \newcommand{\Aufgabe}[2]{\stepcounter{n}
 \textbf{Exercise \arabic{n}: #1} (#2 Punkte)\\}
 
+\newcommand{\Normal}[3]{\mathcal{N}\left(#1,#2,#3\right)}
+\newcommand{\Normalf}[3]{\frac{1}{#3 \cdot\sqrt{2\pi}}e^{-\frac{\left(#1 - #2 \right)^2}{2 #3 ^2}}}
+\newcommand{\Normalfs}[3]{\frac{1}{#3}e^{-\frac{\left(#1 - #2 \right)^2}{2 #3 ^2}}}
+
 
 
 
 \begin{document}
     %\header{BlattNr}{Tutor}{Abgabedatum}{Vorlesungsname}{Namen}{Semester}{Anzahl Aufgaben}
-    \header{7}{}{2015-06-16}{Mobile Robots}{
+    \header{8}{}{2015-06-23}{Mobile Robots}{
         \textit{Jan-Peter Hohloch}\\ \textit{Maximus Mutschler}
-    }{SS 15}{3}
+    }{SS 15}{2}
     \vspace{1cm}
 
     \Aufgabe{Excercise 1}{10}
-    TODO Abbildungen\\
-    d) Measurement uncertainty $Q\rightarrow0
-    \\ K_t = \overline{\Sigma}_tC^TC^{-1}\overline{\Sigma}_t^{-1}(C^T)^{-1}\\
-    =C^{-1} vllt \rightarrow\\
-    \Sigma_t=(I-I)*\overline{\Sigma}_t=0\\
-$ 
-\Aufgabe{Excercise2}{8}    
+    \includegraphics[width=\textwidth]{localize.png}\\
+    %\lstinputlisting{localize_jp/motion_diff.m}
+    %\lstinputlisting{localize_jp/measurement.m}
+    d)
+        The uncertainty in one direction decreases. This causes the estimated trajectory to be quite wiggly (overfitting). We overtrust the measurement.\\
+        For $Q_t$ near to $0$, we get:
+        \begin{align*}
+            K_t&=\overline{\Sigma}_tC^T\left(C\overline{\Sigma}_tC^T+0\right)^{-1}\\
+            &=C^{-1}\\
+            \mu_t&=\overline{\mu}_t+C^{-1}z_t-\overline{\mu}_t\\
+            &=C^{-1}z_t\\
+            \Sigma_t&=0
+        \end{align*}
+        That is we assume no variance and fully trust the measurement.\pagebreak\\
+\Aufgabe{Excercise2}{8}
 \begin{enumerate}[(a)]
-	\item $p(x_t|z_{t,\dots1},u_{t,\dots1})\sim N(x_t,\mu_{x,t},\sigma_{x,t})\\
-	\mu_{x,t}= a\mu_{x,t-1}+bu_t+\epsilon_t\\
-	\sigma_{x,t}= |a|\sigma_{x,t-1}\\
-	\\
-	p(x_{t}|z_{t,\dots1},u_{t,\dots1})\sim N(x_{t-1},\mu_{x,t-1},\sigma_{x,t-1})\\
-	\mu_{x,t-1}= a\mu_{x,t-2}+bu_{t-1}+\epsilon_{t-1}\\
-	\sigma_{x,t-1}= |a|\sigma_{x,t-2}\\
-	\\
-	p(z_t|x-t)\sim N(z_t,\mu_z,\sigma_z)\\
-	\mu_z=c*\mu_{x,t-1}+\delta_t \\
-	\sigma_z=|c|\sigma_{x,t-1}
-	$
+    \item %TODO Herleitung
+        \begin{itemize}
+            \item $p(x_{t-1})=\mathcal{N}\left(x_{t-1},\hat{x}_{t-1},\sigma_{t-1}\right)$
+            \item $p(z_t)=\mathcal{N}\left(z_t,cx_t,r_t\right)$
+            \item $p(x_t)=\mathcal{N}\left(x_t,a\mu_{t-1},q_t\right)$
+        \end{itemize}
+    \item \begin{align*}
+            \Normal{x_t}{ax_{t-1}}{q_t}&=\eta\Normal{z_t}{cx_t}{r_t}\Normal{x_{t-1}}{\hat{x}_{t-1}}{\sigma_{t-1}}\\
+            \Normalf{x_t}{ax_{t-1}}{q_t}&=\eta\Normalf{z_t}{cx_t}{r_t}\Normalf{x_{t-1}}{\hat{x}_{t-1}}{\sigma_{t-1}}\\
+            \Normalfs{x_t}{ax_{t-1}}{q_t}&=\eta\Normalfs{z_t}{cx_t}{r_t}\Normalfs{x_{t-1}}{\hat{x}_{t-1}}{\sigma_{t-1}}
+        \end{align*}
+        %TODO
+	% \item $p(x_t|z_{t,\dots1},u_{t,\dots1})\sim N(x_t,\mu_{x,t},\sigma_{x,t})\\
+	% \mu_{x,t}= a\mu_{x,t-1}+bu_t+\epsilon_t\\
+	% \sigma_{x,t}= |a|\sigma_{x,t-1}\\
+	% \\
+	% p(x_{t}|z_{t,\dots1},u_{t,\dots1})\sim N(x_{t-1},\mu_{x,t-1},\sigma_{x,t-1})\\
+	% \mu_{x,t-1}= a\mu_{x,t-2}+bu_{t-1}+\epsilon_{t-1}\\
+	% \sigma_{x,t-1}= |a|\sigma_{x,t-2}\\
+	% \\
+	% p(z_t|x-t)\sim N(z_t,\mu_z,\sigma_z)\\
+	% \mu_z=c*\mu_{x,t-1}+\delta_t \\
+	% \sigma_z=|c|\sigma_{x,t-1}
+	% $
 \end{enumerate}
-    
+
 \end{document}