20
2

I don't believe these questions are covered by the existing clarification...

(1) Say you are off the road, and there are two grey squares equally close to the goal, but neither is on the branch of the road you were originally traveling on when you left the road. What is the policy supposed to be in this situation?

This could happen under the existing policy by the following sequence of actions from the start:

[down, right, right, right, right, right, right]

So from this white square, both 'up' and 'right' are grey squares you could move to. But neither is on the branch of the road you were originally on - assuming that it's agreed that the grey square directly below the start (which is where you got off the road) is not part of either the branch between the start and the goal, or the branch going directly down from the goal.

(2) Consider the following sequence of actions from the start:

[right, right, right, right, right, right, down, right, left]

So you've moved properly to the right at first, but make a mistake and move down into the off-road white square. Then when trying to move back 'up' to the grey on-road square, you make another mistake and move 'right' to the other grey on-road square. Then when trying to move 'up' into the goal square, you mistakenly move 'left' back into the white off-road square you were just in.

The policy says

If there are two road squares equally close to the goal, head towards the one that is on the branch of the road you were originally traveling on when you first left the road.

When you left the road for the very first time, you were traveling on the upper branch connecting the start to the goal, so the policy would seem to say that you should move 'up'. But when you most recently left the road for the second time, you were traveling on the lower branch going down from the goal, so one would think the actual intent of the policy is that you should try to move 'right'.

So is the intent of the policy actually to always go back to the branch you were traveling on when you first left the road, or is it really to go back to the branch you were traveling on when you most recently left the road?

asked 11 Nov '11, 17:36

jja624's gravatar image

jja624
36118

edited 14 Nov '11, 06:43

aroberge's gravatar image

aroberge
4.0k517

Condition 1) your conclusion is wrong; surely from the start point, there are two distinct branches leading to the goal? In that case, at the final white square the branch you were on when you left the road is the lower branch. Of-course we now have a new clarification that changes the behaviour of the agent, but with the previous clarification the agent should have gone east... no question about it! I hadn't considered your second situation. In this case, the policy would still be to go north as you were on the north branch when you "first" left the road. All bets are off now as we now have the simple "head north" policy which removes all ambiguity. Interesting discussions from the ambiguous question, but with this final clarification it all seems a bit of a waste of our time. It would have been better to stick with the previous clarification which was solvable given the information, but perhaps not providing the answer the tutors originally intended.
(13 Nov '11, 08:30) pmac pmac's gravatar image

Here's a clarification from the Facebook group https://www.facebook.com/groups/242894285747091/

alt text

link

answered 12 Nov '11, 10:27

aroberge's gravatar image

aroberge
4.0k517

2
Great job. It completely changes my answer
(12 Nov '11, 10:36) Oleh Oleh's gravatar image
It changed mine too! I thought it was only fair to share.
(12 Nov '11, 10:38) aroberge aroberge's gravatar image
@aroberge - many thanks! This answers both of my questions.
(12 Nov '11, 10:51) jja624 jja624's gravatar image
1
Thanks for sharing this, @aroberge. It actually goes with my more recent decision about what he was trying to ask, but it's good to know for sure. Anyone else being turned into a Norvig cyber-stalker by this class and the seemingly random locations of clarifications?
(12 Nov '11, 15:41) L_McLean ♦ L_McLean's gravatar image
6
this course becomes more farcical by the minute. The rule was that twitter should be used for official comments, now stuff is being buried in FaecesBook ... appalling behaviour when the staff break the honor code!!!
(12 Nov '11, 21:07) Merlin Merlin's gravatar image
There is a facebbok thing and twitter feed for this class? Hmm... didn't know that.
(12 Nov '11, 21:51) error411 error411's gravatar image
There has been a clarification posted for Homework 5.3: Updated clarification to Homework 5.3 Irvin- 3 hours ago The actions are move North, West, East, and South. All actions are stochastic; 80% they move as intended, and 10% they might move 90 degrees right or left. The part of the policy c) of moving back immediately means on the next turn take an action that (if it goes in the intended direction) brings the agent back to the grey square that is closest to its position, and if there are several of those, closest to the goal. If there are two road squares equally close to the goal, head North. https://www.ai-class.com/course/video/quizquestion/167/1
(13 Nov '11, 02:23) robrambusch ♦ robrambusch's gravatar image
These clarification isn't mentioned on the question page. Should I use it or no?
(16 Nov '11, 09:44) egens egens's gravatar image
@egens: use what Peter Norvig says last in the screencapped Facebook discussion above, which matches what they finally put on the question page.
(16 Nov '11, 12:25) L_McLean ♦ L_McLean's gravatar image
thanks a lot :)
(16 Nov '11, 13:20) AngieB AngieB's gravatar image
@L_mclean but it doesn't match to the info that is on the question page right now. in words that we must head in back in the direction we came from when went offroad - there is no such clarification.
(16 Nov '11, 15:35) egens egens's gravatar image
1
@egens -- use what's on the homework page. The "back to the branch you were on when you first went offroad" used to be on the homework page but was replaced with "head North" because it was still unclear. What's now posted on the homework page should be 100% clear for all cases that we need to worry about.
(16 Nov '11, 15:41) L_McLean ♦ L_McLean's gravatar image
@L_mclean I red the words "head North", but it was still unclear for me, because I thought these words are for the gray road states. Now I understand that this rule can be used when agent is off the road. Thanks.
(17 Nov '11, 02:31) egens egens's gravatar image

I don't believe you follow the rules of the game. The first rule (rile a) is to go to the square (on a road) that is closest to the goal. In your first example, you do not do that. Your first move is down, when it should be right.

Am I missing something?

Actually, there is a significant ambiguity in the following case:

a1 intend Right, but actually goes Down

a2 Intend up, but actually goes Right

Now, ambiguity: do I go left or do I go up?

Answer to this question changes drastically the answer to the Homework,I believe... (By the way, my answer is to go up--back to the branch I intended to be on...)

link

answered 11 Nov '11, 20:06

OldieButGoodie's gravatar image

OldieButGoodie
1.6k314

edited 12 Nov '11, 02:34

"All actions are stochastic; 80% they move as intended, and 10% they might move 90 degrees right or left."
(12 Nov '11, 01:58) duffahtolla duffahtolla's gravatar image
1
I understand But... The first action is down. Two cases: You intended to go down, but then break rule number 1 (you are not going toward the goal) You intended to go right but went down. In this case, your next action should be to go up. But then you went right At this point you have two choices: go left or go up. If you intend to go left, you cannot stochastically go right. So that was not what you intended to do. What you intended to do was go up--and instead you ended up going right. But then, there is no ambiguity at the end. You should intend to go up... Ok?
(12 Nov '11, 02:40) OldieButGoodie OldieButGoodie's gravatar image
The scenario I am mapping out in my first question is your no. 2: you intended to go right but went down. Then you intended to go up but went right (into a white square). But now at each of the next five stages you intended to go up into a grey square (about which there is no ambiguity using the stated policy), but end up going right. The only ambiguity is when you get to the final white square, and the policy doesn't say whether you should be going up or right.
(12 Nov '11, 07:13) jja624 jja624's gravatar image
When you get to the final white square, the policy does say where to go, in my opinion.
(12 Nov '11, 08:18) dlask ♦ dlask's gravatar image
@dlask: for the first example I gave above, I disagree. At the final white square, both up and right are road squares that are equally close to the goal. At this point, the policy says to choose the road square that is on the branch of the road you were originally traveling on when you first left the road. But the branch you were traveling on when you first left the road is the one going down from the start, and neither the up nor the right squares are on this branch. (Unless you have a different definition of branch than I do - but feel free to describe it...) For the second example I gave above, the policy does say where to go, but it would say to go up instead of right, even though you most recently came off the road from the right rather than from up. So although the policy does give a direction, it seems to give a direction that is different from what I think was supposed to be the intent of the policy. In any case this is all moot since the facebook discussion below answers both of my questions.
(12 Nov '11, 10:49) jja624 jja624's gravatar image
@OldieButGoodie I understand your situation, even with the latest notes on the problem. Being at S you have to go right (stay gray && minimize dist G) but you end up at c1 (a-e and 1-8 nomenclature) going down, now we are in grey so the next is to go up (minim dist to G) but we end up in c2. Now we can go a) up (b2) to minimize dist G or b) go back to where we came from (c1). If we take b) then many other things can happen differently from option a). Is policy over going back to road from where we left of or policy over mini. dist. G more important?
(13 Nov '11, 14:43) casathom casathom's gravatar image

I actually think the problem is more interesting when the policy while on a white square is "try to rejoin the road you most recently left" and it does change the answer to have a special policy of "up" for the rightmost infield white square. Of course I'll go with Prof. Norvig's policy.

link

answered 12 Nov '11, 22:14

RunDMC's gravatar image

RunDMC
562

I completely agree with you on this one. The previous policy seemed much more elegant, and should have been amended adding a random selection in the case of draw of the "back to the branch you were traveling on when you most recently left the road". I thought it was sweet.
(13 Nov '11, 05:42) Bonghi Bonghi's gravatar image

Heading towards
  the goal station
my route is a
  policy violation.

link

answered 12 Nov '11, 03:26

dlask's gravatar image

dlask ♦
9.7k218

(12 Nov '11, 03:48) aaronnewton aaronnewton's gravatar image

I wasn't confused until I read this.

My assumption was that if you went off the road south just before arriving at G and then tried to go north to rejoin the road, but went east instead. You have none the less rejoined a road and should make for the goal G by heading north. If this results in going back west, back into the square you just left, then the road to your east was the road you were last on and therefore you should try to go east again. This is how I interpreted the last line of the clarification. Am I now to assume that rejoining a road stochastically doesn't count and I should try to go back off the road again, risking stochastically arriving at G?

link

answered 12 Nov '11, 18:50

Derek's gravatar image

Derek
313

1
From Peter Norvig's last comment in the Facebook conversation that @aroberge posted above: Let's agree with Curt and just say that the policy from that square is "up". Gosh I wish he had just said that to begin with ;)
(12 Nov '11, 19:00) L_McLean ♦ L_McLean's gravatar image
1
Actually, that is what I did before I read this post. I think the clarification was pretty clear on the homework page. The Facebook message is just give away answer. I also wished he could be consistent and gave away the same answer in the homework clarification :)
(12 Nov '11, 20:14) slee slee's gravatar image

Don't get caught up in the details, since the question asks you to click the squares the agent won't explore. I think the question has enough in it to get it right. I think Peter Norvig is just being polite in his facebook clarifications.

link

answered 13 Nov '11, 01:22

dougfinn's gravatar image

dougfinn
5.4k627

edited 13 Nov '11, 01:35

I could not edit it, so I deleted a comment which added specific information about the effect of the clarification. The comment was by @Michael and read partly as follows: Incorrect, new policy of going up --deleted--. You actually MUST get all the details right, for example I wrote the program to calculate the values and the output and this addendum changes it quite dramatically.
(14 Nov '11, 09:52) aroberge aroberge's gravatar image
1
Sorry I did not realize I was too specific. Incorrect, new policy of going up changes answer. You actually MUST get all the details right, for example I wrote the program to calculate the values and the output and this addendum changes it quite dramatically.
(14 Nov '11, 09:54) Michael Michael's gravatar image

I would say it's fairly easy to see what the instructors intended even if you can find an example where they didn't cover every circumstance. I understand the frustration when the directions aren't exhaustive, especially considering the fact that this is a programing class where we have to tell a computer exactly what to do. I would also be quite upset if I got question wrong because I misinterpreted what they where asking. However, we aren't computers who can't fill in the gaps, so if they don't further clarify I would suggest relying on your intuition of what they want.

link

answered 11 Nov '11, 18:58

Mike%20H's gravatar image

Mike H
2.8k112

2
As you can see from the fact that people (including myself) changed their answers after getting this clarification from the professor, relying on intuition didn't help here :) If only the professors / assistants would be so kind as to either (a) put these answer-changing clarifications in the problem spec; (b) really think all these details through before posting problems in the first place.
(12 Nov '11, 11:05) jja624 jja624's gravatar image

For what it's worth, I came here for exactly the same question, and have decided to read the final part of the clarification:-

"head towards the one that is on the branch of the road you were originally traveling on when you first left the road."

as

"head towards the one that is on the intended route you were originally traveling on when you first left the road.

If anyone thinks differently, please let us know.

link

answered 12 Nov '11, 10:28

Rich%20P's gravatar image

Rich P
11

I do. Ambiguity arises from understanding "the branch of the road", that equals to your "intended route". I could call a horizontal edge of gray square one route, and the rest three edges would combine into other route, which would be intended if policy says so. But aroberge has already shed light on that question.
(12 Nov '11, 10:45) Oleh Oleh's gravatar image
It makes no sense to "read" the clarification in a different way. Surely the agent has to follow the rule to the word, so in that case it is very clear that the rule specifies heading towards the branch (there are two branches leading to the goal in this example) that you were on when you first left the road (road being the grey squares). No-where does the policy specify "indented route" (which I would translate as intended BRANCH to be more specific). This is all moot now we have a new clarification which contradicts the previous clarification; the new behaviour to head north would not be possible with the previous policy clarification given the condition that the agent goes south in the first step before leaving the road...
(13 Nov '11, 09:00) pmac pmac's gravatar image
The original question in the thread asked where you should go if you originally left the road from the square just south of the start, and haven't been back to a road at any time after that. Neither adjacent road cell is on the branch you left from but it could be read as the 'branch that is on the shortest route from where you left the road' or 'branch that is on the route you were actually, though unintentionally, travelling when you left the road. These possibilities produce a very different set of results
(14 Nov '11, 04:43) Rich P Rich%20P's gravatar image

I think the most correct policy (in the sense that it is apparently what Prof Norvig intended) is "toward the road you ought to have been on the very first time you ever exited, if your agent hadn't stochastically screwed up even earlier than that!" ;^) [Not knowable by the agent, of course!]

link

answered 12 Nov '11, 17:45

Dudley%20Brooks's gravatar image

Dudley Brooks
67110

edited 12 Nov '11, 17:56

What gets me is the use of Right/Left and 90 degress from intended direction. If the agent is moving right to get to the goal, then stochastically it can end up in either north or south cell. Let's say the agent stochastically moves south instead of right (90 degress form intended direction), then when it trys to move north to get back on the gray road, it then stochastically moves right. When they say it can stochastically move right or left, then it will always be a 10% propability it will move toward the goal. Which is it Right/Left or 90 degrees form intended direction?? I lean toward 90 degrees from intended direction.

link

answered 12 Nov '11, 22:32

mijs's gravatar image

mijs
2015

Definitely 90 degrees from intended direction. Rewatch the video and pay close attention when he's demonstrating where the agent could possibly end up after the first move.
(12 Nov '11, 22:39) L_McLean ♦ L_McLean's gravatar image

I attemped to touch upon this at my blogs. http://mostafaabedi.blogspot.com or http://mostafaabedi.wordpress.com

*I have modified my blogs

link

answered 13 Nov '11, 02:13

mostafaabedi's gravatar image

mostafaabedi
2785

edited 13 Nov '11, 02:15

Also, go back to the other rule... always (try to) move closer to the goal. Remember your geometry.

link

answered 13 Nov '11, 04:55

japina's gravatar image

japina
612

Your answer
toggle preview

Follow this Question via Email

Once you sign in you will be able to subscribe for any updates here

Q&A Editor Basics

  • to upload an image into your question or answer hit
  • to create bulleted or numbered lists hit or
  • to add a title or header hit
  • to section your text hit
  • to make a link clickable, surround it with <a> and </a> (for example, <a>www.google.com</a>)
  • basic HTML tags are also supported (for those who know a bit of HTML)
  • To insert an EQUATION you can use LaTeX. (backslash \ has to be escaped, so in your LaTeX code you have to replace \ with \\). You can see more examples and info here

powered by OSQA