t test for intercept? the null hypothesis for slope is usually H0: slope is zero,but what's thenull hypothesis for intercept? is it H0:intercept is zero as well?
From the picture we can see that the p value for intercept is large, so we don't reject the null hypothesis, which means that the null hypothesis for intercept should be H0: intercept is non-zero?

 A: Hypothesis testing is like mathematical ''proof by contradiction''; if you want to prove something, then you assume that the opposite is true and using this ''opposite is true'' assumption you try to find a contradiction.  As contradictions are impossible, the assumption ''opposite is true'' must be false.  
In hypothesis testing you do the same; if you want to show that the intercept (or the slope) is signficantly different from zero, then you assume the opposite, i.e. $H_0: \beta_0 = 0$ and try to derive a contradiction from this. As in statistics nothing is impossible we will not be able to derive something ''contradictory'' but we will try to show that this leads to something ''very improbable''.  
So you first have to define what you mean by ''very improbable'' i.e. you must chose a siginficance level e.g. 5%. If the probability of ''something'' is below this chosen significance level we will take it as very ''improbable''.
To summarize (a) define what you mean by ''very improbable'', i.e. define your significance level e.g. $\alpha=0.05$ (b) if you want to ''show'' that $\beta_0 \ne 0$ then assume to opposite (i.e. $H_0: \beta_0 = 0$ and try to find a ''contradiction'' i.e. something that occurs with low probability. 
The theory of linear regression learns that, if $H_0: \beta_0=0$ is true (and the assumption of linear regression are  fulfilled), then the estimate of $\beta_0$, i.e. $\hat{\beta_0}$ has a normal distribution with mean $\beta_0$, but the latter we assumed it to be equal to zero.  The p-values in your table are derived from this normal distribution. It seems to be 0.4 which is not below our chosen significance level $\alpha=0.05$, so we do not find a ''contradiction'' and we can not ''prove'' that $\beta_0 \ne 0$ 
Note;  if we say that we can not prove that $\beta_0 \ne 0$ then this does in no way mean that we can prove that it is zero !
So if we reject the null hypothesis then we have ''statistically proven'' $H_1$, if we can not reject $H_0$ then we can not conclude that we can prove $H_0$, we simply accept it (which is different from having proven it). 
Applying this to e.g. the coeffient of 'Horsepower', let me call this coefficient $\beta_1$: if I want to show that this coefficient is different from zero then I assume the opposite: $H_0: \beta_1 = 0$ and assuming this I can find a p-value of zero, so someting very improbable.  So the assumption $H_0: \beta_1 = 0$ leads to something that is very improbable, therefore it must be false and thus $H_1: \beta_1 \ne 0$ is ''statistically proven''. 
Similar for the intercept, let me call it $\beta_0$; if we want to show that it is non-zero, then assume the opposite $H_0: \beta_0=0$ , if that is true then you found a p-value of 0.4, not so ''improbable'' thus.  So the assumption $H_0: \beta_0 = 0$ does not lead to a ''statistical contradiction'' and we find no evidence that $H_1: \beta_0 \ne 0$ is true. 
Note that 'finding no evidence that $H_1:\beta_0 \ne 0$ is true' does not imply that the opposite ($H_0: \beta_0 = 0$)  is ''statistically proven'', we simply can not find indications that $H_0$ could be false, therefore we ''accept it''. 
Important remark:  the assumption $H_0: \beta_0 = 0$ is used in the computation of the p-value.  It is only because we assume that $\beta_0 = 0$ that we can compute a p-value, without that assumption we do not know the distribution and can not compute probabilities. 
So - in the same logic as above - if I want to ''statistically prove'' that $\beta_0 = 0$ then I have to assume the opposite, so my $H_0$ would then be: $H_0: \beta_0 \ne 0$ and we have to find something improbable so that I can reject $H_0$.  But there is a problem here,  when I assume that $H_0: \beta_0 \ne 0$ then I can not guess the value of $\beta_0$ (it could be anything that is not zero, so 5 or 1 million, ...).  So my assumption $H_0: \beta_0 \ne 0$ does not allow me to fix the distribution and does not allow me to compute p-values ... so I am stuck here. 
Very brief summary: the goal is to reject $H_0$ in order to find ''statistical evidence'' for $H_1$, when $H_0$ can not be rejected we just ''accept'' it. The assumption under $H_0$ must be precise enough to fully determine the distribution of the test statistic, else we can not compute p-values. 
A: It is also $H_0:\beta_{0,0}=0$ in your example. You can infer that from the general formulation of a t-ratio
$$
t=\frac{\hat\beta_j-\beta_{j,0}}{std.error(\hat\beta_j)},
$$
where $\beta_{j,0}$ is the hypothesis formulated on $\beta_j$ and $\beta_0$ is the coefficient on the intercept.
In your case, we have 
$$
.837=\frac{.128}{.154},
$$
so nothing is subtracted from $\hat\beta_0$ in the numerator, thus $\beta_{0,0}=0$. But as the above hopefully makes clear, you could test any hypothesis you happen to be interested in using the coefficient estimate .128 and the standard error .154 by formulating your own t-ratio. 0 is just the default reported by the package.
