Could someone show me some key things to keep in mind while grabbing pawns?

Here's Hikaru.

Check out 2:19.

He is in a position where he's able to grab pawns (just entering the middle-game). He's given the option of a free central pawn, and then a free wing pawn, too.

He takes the free central pawn (valuable!), but declines to take the free wing pawn (not so valuable).

Sometimes too much pawn-grabbing can destabilize your position. You've got a whole army to deal with. Don't spend too much time chasing lowly foot soldiers ...

Great tips in here! 

I feel the advice so far is not completely to my liking. In general you should want to take a free pawn. Usually it's up to your opponent to prove their compensation is sufficient and such compensation isn't often worth more than a pawn. If their compensation dies out, you'll have clear winning chances. Of course, this does come in degrees and it's extremely hard to generalize. 2-3 extra developing moves might be worth a pawn in the opening, but probably not in a closed position and maybe not if their king is safe. Also, one might have more trouble in a blitz game to nullify their compensation than in a full 2hr game. So it really depends on that too. That being said, quite often leaving a pawn to be taken is a very agressive way of playing by your opponent. If you don't take the pawn, they'll have all of their advantages from the aggressive move and none of the downsides if you don't just take it...

The 1 general rule that might be good to know is to 'never take on b2 or b7 even if it's right'. That has been attributed to a strong player. Now this might be a little bit of an exaggeration, but it does often hold some truth to it. Especially because the queen might get trapped. Only calculation will have to determine in a specific position whether you can and should take it anyway. I suggest looking at puzzles trapping queens on b2 or b7 to get better at spotting the potential trouble. And looking at dynamic games is a good way to learn about the dynamics in chess and when it's worth a pawn. 

A pawn is a pawn. If you can, grab the pawn, defend and win. To refute a gambit, accept it.
Do not get greedy. 1 pawn is enough to win. It is often unwise to grab a 2nd or 3rd pawn.

As I've said, even very strong players can be unsure if they should take b2/b7.

For instance Kasparov played poisoned pawn variation of the Najdorf as black, where white has a lot of initiative afterwards.

On the other hand, I've heard (I am not sure where, maybe in Chess Dojo videos) that he disliked getting into Marshall attack of the Ruy Lopez as white because black gets a lot of initiative, and it doesn't suit his style to hand the initiative to the opponent. For me as a patzer, it seems that in both positions the side with a pawn down position has the initiative, so I am not sure why he liked playing poisoned pawn variation and he didn't like Marshall gambit.

That’s interesting. Thank you. I wish he would explain more as to why, however. But, even I wouldn’t take it. Opens the h-file, maybe for an attack. 

Very informative. Yes, I should check out the puzzles. Thank you. 

Thank you. I will try to remember all this advice in my games. 

That’s strange. Taste, I guess. 

Oh I wouldn't call yourself a patzer...

It might have something to do with the winning chances. Playing as black in the Poisoned Pawn there's a lot of the game still to play. The Marshall occurs a lot later. And maybe he had analysed it and just couldn't find anything suitably challenging for black, leading to higher drawing chances if he would play against it. The Marshall is still formidable at top GM level. Besides, with white you have different goals than with black, who usually plays for a draw at the highest level. Adding a chance to win with black might have suited his taste a bit more. 

Thank you for the kind words. 

You are probably right about the poisoned pawn (and Marshall), I tried it once in a daily game as black and I will not try it again, at least not in the near future.

It is really interesting how some positions are viewed differently through history. 20 years ago Berlin was played for a draw, at certain point Marshall was considered as a good try to win. Now Marshall has been played even more along with Berlin, and I've heard that on super GM level Marshall is now played for a draw and Berlin for a win.

Great question, Aun.  Here's a question that's the flip side: when is *allowing* the taking of a pawn ok?  I've been experimenting with French, and I'm scared to play the Winower because some of the main lines has (if I remember right) white swooping in with a queen to take the g7 pawn, and the books say: "this is ok for black, white doesn't really have a threat".  (My response is: "um, really?  it sure looks like a threat!"  And sometimes it is a poisoned pawn.

Since I can't tell when/which -- I never play it.

(If anyone has a good example, I invite them to post it here!)

These kinds of things tend to come down to study.

Strong players usually don't wander into poisoned pawn variations (from either side). It's most often something they've studied and prepared for.

So it's not so much a matter of figuring out what to do ... it's more a matter of remembering your prep, and understanding, from experience, what to look out for (and/or knowing how the line should go).

What Stil says is certainly the truth for higher rated players. For us, we have to calculate, and see what will most likely happen within next few moves. It is important to have an idea how to play after sacrificing that pawn.

Here is a game where I could sacrifice that pawn, but I didn't:

Instead of playing Nc3 and sacrificing the pawn, I went for dxc5 which is surely not right, even I sort of knew it during the game.

The idea for playing 8. Nc3 is rougly this: Black lacks development, and if he takes the pawn I play Nb5 threatening Nd7+ winning a rook, but it is complex afterwards to be fair. For instance if he plays c4, he will sacrifice the exchange (if I play immediate Nd7+), but he will win c2 pawn as well. So black would have 2 pawns for the exchange and protected passed pawn on c4. While I take the rook, his development would probably catch up. So Nd7+ there is not the best, it is probably more accurate to play Rb1 first.  By the way I say the exchange... White will actually by up a rook for some pawns, but knight on a8 is out of the game, and most likely lost.

Apart from that he doesn't have to play c4, he could play Bxc2 immediately after my Nb5 move, so even though it really looks like white is better if black takes the pawn (and white is probably better), it is not easy to exploit this if black knows what he is doing afterwards.

In short, you have to decide during the game. Myself, I hate letting go of the pawn and I have tendencies to keep the material as much as I can, but that is a weak player mindset talking.

Here I got away with it, but I sometimes lose games because of that mindset.

I'm surprised no one told you to look up for some famous games in poisoned pawn variation of any opening (on wikipedia they list five of them, but I think Sicilian Defense, Najdorf Variation is definitely most famous). Agadmator probably has plenty of videos covering posioned pawn variations and there are wins for both sides - the one who offers a pawn and the one who takes it. Even better is to read annotations it gives you bigger picture especially if game is annotated by a player who played it himself. Just find any strong player that likes to play either side and look up few of their games. I remember seeing a game on someones blog and wondering why the guy didn't take free pawn, maybe there was some trap I couldn't see. After asking that question he answered something among the lines its totally fine to take the pawn but I haven't even thought of it during the game. His priorities were developing, castling etc. Its his way of play and its neither right or wrong. Doing this will help you see most common replies, what is cosidered good and what is not and you will get yourself more familiar with positions that can arise from taking that pawn. In your games always analyze was the way you played the only possible way. Opportunity to take poisoned pawn is almost always in opening phase of the game or late opening/early middlegame. So my suggestion would also be to check after the game are there any more games with reached position and continuations of them.

But this is also such an interesting question to think about! I hope OP doesn't mind me going off topic along with the people that I'm responding to...

Basically when you give up a pawn, it is because you get other 'advantages'. Usually these advantages are dynamic in nature. So you'd get more active pieces, a lead in development or the infamous initiative. It's stuff we can't count, it might even be something we can't necessarily see on the board. Apart from that, these advantages can also be fleeting. You don't lose a pawn advantage unless they win one back, but if you give your opponent enough time, they'll become just as active as you were, nullifying that dynamic advantage that you had. So basically your goal with the dynamic advantages is (usually) to be quick and to try and gain something with it, either winning material back or getting more dynamic advantages. 

I think there are 2 important challenges that come with playing the side that has sacrificed the pawn for the dynamic advantages:
1. Thinking you need to punch back immediately because you're down material. This I see coming back a bit in the Winawer reply and in the remark 'for us, we need to calculate, and see what will most likely happen within next few moves'. That's not necessarily true. Look at it as a trade: I trade you a (small) material advantage ('you're up a pawn') while you trade me these dynamic advantages ('but my pieces are great'). That can be just as fair a trade as trading a bishop for a knight. So, relax. As long as you have that dynamic advantage, you don't need to go all out trying to win back material immediately. For example, in the Winawer black might have some nice play in the center. It might have cost the h-pawn and that's a problem in the endgame, but you're not there yet! That h-pawn is not about to queen, so you have time to play your trumps in the middlegame. Hopefully the game will never reach the endgame.
2. Not knowing how to play with the dynamic advantages. The initiative is especially hard. Having the initiative is basically being the one that makes the threats while improving his own position, while the other player needs to continuously respond to the threats and can't improve his position. Good dynamic play very often requires you to keep finding new problems for your opponent, keep making threats at them. In this sense, it's linked with a strong understanding of static play (doubled pawns, outposts, open files, etc.). A simple example: maybe I can develop my bishop to attack his knight. I threaten to take the knight and he'll get doubled pawns. I need to know whether this is a real threat or not. If he can just ignore it and make a much needed developing move of his own, that hurts my dynamic advantages. However, if he has to defend against the threat by retreating his queen to guard the knight, I just made a developing move with tempo adding to my dynamic advantages. Good understanding of static play means you don't only have tactical threats you can use, but you can also use strategic threats to increase your advantage. Another example: if I have a lead in development, it might mean I can place both of my rooks on the only open file. If I don't know how to use an open file, because that part of my chess understanding is not (yet) there, I might wait too long and my opponent again could be just in time to defend against my rooks coming down the open file. Basically, playing dynamic chess means making your moves count and knowing what you're doing. 

I've always thought that understanding dynamic chess is hard. Even as a 2000 player, I didn't really understand it. And because you need so much understanding of static chess advantages too, I've always thought that it's much better for advancing players to focus on really understanding the static characteristics of chess positions first before even attempting to play dynamic chess. So if you don't feel ready to go sacrificing pawns yet, I hear you! Then again, it can't hurt to just try it out once in a while.

In a sense this is all on topic. The question was when to take the pawn, but it is reasonable enough to discuss when to sacrifice it, it complements the question rather nicely. And I am sure everyone else also appreciates when a stronger player offers his advice on some finer points of the game.

Yeah, you are of course correct. You can''t just calculate everything. In the game I've linked, it is safe to say that you can't calculate everything. It is just that you will probably play Nc3 with the idea of Nb5 afterwards. You will see that black is lagging in development, you have some threats and can say: fine, this looks good enough, he shouldn't be able to be better in that position even a pawn up and just go from there.

But of course when you can calculate, why not. In many cases I consider the top engine move, my gut feeling says this is the best, but then when I calculate it seems that it just doesn't work. In some cases I figure out that it does work, I just couldn't find my response to what I was afraid of.

As for dynamic play, in many cases I don't go for the sacrifice as I am not sure and don't want to just end up with material deficit. Sometimes I feel that I am much better with handling straightforward and easily valued material, than evaluating compensation. But then in some game I lose a pawn anyways and then find a way to make threats and salvage the game. So when I am forced to play for activity vs material, sometimes I actually surprise myself.

To be fair, I think it all comes down to experience and you get better at it the more you play and analyze your games.

Thank you all for the responses! I learned a lot!

Same. I am always scared to sacrifice pawns because I am not good at using dynamic advantages and calculating complicated positions.