Project 1: Search in Pacman

Introduction

In this project, your Pacman agent will find paths through his maze world, both to reach a particular location and to collect food efficiently. You will build general search algorithms and apply them to Pacman scenarios.

As in all future projects, this project includes an autograder for you to grade your answers on your machine. This can be run with the command:

python autograder.py

See more on using the autograder, run python autograder.py -h

The code for this project consists of several Python files, some of which you will need to read and understand in

order to complete the assignment, and some of which you can ignore. You are provided all the code and supporting files in the zip of this assignment.

Files you'll edit:

search.py              Where all of your search algorithms will reside.

searchAgents.py        Where all of your search-based agents will reside.

Files you might want to look at:

pacman.py          The main file that runs Pacman games. This file describes a Pacman GameState type, which you use in this project.

game.py         The logic behind how the Pacman world works. This file describes several supporting types like AgentState, Agent, Direction, and Grid.

util.py           Useful data structures for implementing search algorithms.

Supporting files you can ignore:

Files to Edit and Submit: You will fill in portions of search.py and searchAgents.py during the assignment. You should submit these files with your code and comments. Please do not change the other files in this distribution or submit any of our original files other than these files. You will not need to zip your submission.

Evaluation: Your code will be autograded for technical correctness, using the same autograder and test cases you are provided with. Please do not change the names of any provided functions or classes within the code, or you will wreak havoc on the autograder. You should ensure your code passes all the test cases before submitting the solution, as we will not give any points for any questions if not all the test cases for it pass. However , the correctness of your implementation -- not the autograder's judgements -- will be the final judge of your score .

Even if your code passes the autograder, we reserve the right to check it for mistakes in implementation, though this should only be a problem if your code takes too long or you disregarded announcements regarding the project.

Academic Dishonesty: We will be checking your code against other submissions in the class for logical redundancy. If you copy someone else's code and submit it with minor changes, we will know. These cheat detectors are quite hard to fool, so please don't try. We trust you all to submit your own work only; please don't let us down. Likewise, do not attempt to write your code specifically to pass the autograder's tests. Either copying or trying to cheat the autograder will be considered violations of the student honor code.

Getting Help: You are not alone! If you find yourself stuck on something, contact the course staff for help.

Office hours and Piazza are there for your support; please use them. If you can't make our office hours, let us know and we will schedule more. We want these projects to be rewarding and instructional, not frustrating and demoralizing. But, we don't know when or how to help unless you ask.

Imports: Do NOT import any new libraries for this project (ie. numpy, scikit learn, pandas, etc). Some of these libraries trivialize the assignment and importantly they are not installed on the gradescope autograder. Importing  these libraries will crash the autograder and that submission will recieve a zero. Please also check that your IDE did not incidentally import libraries (common with VSCode for instance). See more in the debugging gradescope section.

Welcome to Pacman

After unzipping the code and changing to its directory, you should be able to play a game of Pacman by typing the following at the command line:

python pacman.py

Pacman lives in a shiny blue world of twisting corridors and tasty round treats. Navigating this world efficiently will be Pacman's first step in mastering his domain.

The simplest agent in searchAgents.py is called the GoWestAgent, which always goes West (a trivial reflex agent). This agent can occasionally win:

python pacman.py --layout testMaze --pacman GoWestAgent

But, things get ugly for this agent when turning is required:

python pacman.py --layout tinyMaze --pacman GoWestAgent

If Pacman gets stuck, you can exit the game by typing CTRL-c into your terminal.

Soon, your agent will solve not only tinyMaze, but any maze you want.

Note that pacman.py supports a number of options that can each be expressed in a long way (e.g., --layout) or a short way (e.g., -l). You can see the list of all options and their default values via:

python pacman.py -h

Also, all of the commands that appear in this project also appear in commands.txt , for easy copying and pasting. In UNIX/Mac OS X, you can even run all these commands in order with bash commands.txt.

Note: if you get error messages regarding Tkinter, see this page

Object Glossary

Here's a glossary of the key objects in the code base related to search problems, for your reference:

SearchProblem (search.py)

A SearchProblem is an abstract object that represents the state space, successor function, costs, and goal   state of a problem. You will interact with any SearchProblem only through the methods defined at the top

of search.py

PositionSearchProblem (searchAgents.py)

A specific type of SearchProblem that you will be working with --- it corresponds to searching for a single

pellet in a maze .

CornersProblem (searchAgents.py)

A specific type of SearchProblem that you will define --- it corresponds to searching for a path through all four corners of a maze .

FoodSearchProblem (searchAgents.py)

A specific type of SearchProblem that you will be working with --- it corresponds to searching for a way to eat all the pellets in a maze .

Search Function

A search function is a function which takes an instance of SearchProblem as a parameter, runs some algorithm, and returns a sequence of actions that lead to a goal. Example of search functions are

depthFirstSearch and breadthFirstSearch, which you have to write. You are provided

tinyMazeSearch which is a very bad search function that only works correctly on tinyMaze

SearchAgent

SearchAgent is a class which implements an Agent (an object that interacts with the world) and does its planning through a search function. The SearchAgent first uses the search function provided to make a  plan of actions to take to reach the goal state, and then executes the actions one at a time.

Finding a Fixed Food Dot using Search Algorithms

In searchAgents.py, you'll find a fully implemented SearchAgent, which plans out a path through Pacman's world and then executes that path step-by-step. The search algorithms for formulating a plan are not

implemented -- that's your job. As you work through the following questions, you might find it useful to refer to the object glossary (the second to last tab in the navigation bar above) .

First, test that the SearchAgent is working correctly by running:

python pacman.py -l tinyMaze -p SearchAgent -a fn=tinyMazeSearch

The command above tells the SearchAgent to use tinyMazeSearch as its search algorithm, which is implemented in search.py. Pacman should navigate the maze successfully.

Now it's time to write full-fledged generic search functions to help Pacman plan routes! Pseudocode for the

search algorithms you'll write can be found in the lecture slides. Remember that a search node must contain not only a state but also the information necessary to reconstruct the path (plan) which gets to that state.

Important note: All of your search functions need to return a list of actions that will lead the agent from the start to the goal. These actions all have to be legal moves (valid directions, no moving through walls) .

Important note: Make sure to use the Stack, Queue and PriorityQueue data structures provided to you in

util.py! These data structure implementations have particular properties which are required for compatibility with the autograder.

Hint: Each algorithm is very similar. Algorithms for DFS, BFS, UCS, and A* differ only in the details of how the fringe is managed. So, concentrate on getting DFS right and the rest should be relatively straightforward. Indeed,

one possible implementation requires only a single generic search method which is configured with an

algorithm-specific queuing strategy. (Your implementation need not be of this form to receive full credit) .

Question 1 (2 points)

Implement the depth-first search (DFS) algorithm in the depthFirstSearch function in search.py. To make your algorithm complete , write the graph search version of DFS, which avoids expanding any already visited states.

Your code should quickly find a solution for:

python pacman.py -l tinyMaze -p SearchAgent

python pacman.py -l mediumMaze -p SearchAgent

python pacman.py -l bigMaze -z .5 -p SearchAgent

The Pacman board will show an overlay of the states explored, and the order in which they were explored

(brighter red means earlier exploration). Is the exploration order what you would have expected? Does Pacman actually go to all the explored squares on his way to the goal?

Hint: If you use a Stack as your data structure, the solution found by your DFS algorithm for mediumMaze should have a length of 130 (provided you push successors onto the fringe in the order provided by

getSuccessors; you might get 246 if you push them in the reverse order). Is this a least cost solution? If not, think about what depth-first search is doing wrong.

Question 2 (2 points)

Implement the breadth-first search (BFS) algorithm in the breadthFirstSearch function in search.py. Again,    write a graph search algorithm that avoids expanding any already visited states. Test your code the same way you did for depth-first search.

python pacman.py -l mediumMaze -p SearchAgent -a fn=bfs

python pacman.py -l bigMaze -p SearchAgent -a fn=bfs -z .5

Does BFS find a least cost solution? If not, check your implementation.

Hint: If Pacman moves too slowly for you, try the option --frameTime 0.

Note: If you've written your search code generically, your code should work equally well for the eight-puzzle search problem without any changes.

python eightpuzzle.py

Varying the Cost Function

While BFS will find a fewest-actions path to the goal, we might want to find paths that are "best" in other senses . Consider mediumDottedMaze and mediumScaryMaze.

By changing the cost function, we can encourage Pacman to find different paths. For example, we can charge more for dangerous steps in ghost-ridden areas or less for steps in food-rich areas, and a rational Pacman agent should adjust its behavior in response.

Question 3 (2 points) Implement the uniform-cost graph search algorithm in the uniformCostSearch function in search.py. We encourage you to look through util.py for some data structures that may be useful in your

implementation. You should now observe successful behavior in all three of the following layouts, where the agents below are all UCS agents that differ only in the cost function they use (the agents and cost functions are written for you):

python pacman.py -l mediumMaze -p SearchAgent -a fn=ucs

python pacman.py -l mediumDottedMaze -p StayEastSearchAgent

python pacman.py -l mediumScaryMaze -p StayWestSearchAgent

Note: You should get very low and very high path costs for the StayEastSearchAgent and

StayWestSearchAgent respectively, due to their exponential cost functions (see searchAgents.py for details) .

A* search

Question 4 (3 points)

Implement A* graph search in the empty function aStarSearch in search.py. A* takes a heuristic function as   an argument. Heuristics take two arguments: a state in the search problem (the main argument), and the problem itself (for reference information). The nullHeuristic heuristic function in search.py is a trivial example.

You can test your A* implementation on the original problem of finding a path through a maze to a fixed position using the Manhattan distance heuristic (implemented already as manhattanHeuristic in searchAgents.py) .

python pacman.py -l bigMaze -z .5 -p SearchAgent -a fn=astar,heuristic=manhattanHeuristic

You should see that A* finds the optimal solution slightly faster than uniform cost search (about 549 vs. 620

search nodes expanded in our implementation, but ties in priority may make your numbers differ slightly). What happens on openMaze for the various search strategies?

Finding All the Corners

The real power of A* will only be apparent with a more challenging search problem. Now, it's time to formulate a new problem and design a heuristic for it.

In corner mazes , there are four dots, one in each corner. Our new search problem is to find the shortest path

through the maze that touches all four corners (whether the maze actually has food there or not). Note that for

some mazes like tinyCorners , the shortest path does not always go to the closest food first! Hint: the shortest path through tinyCorners takes 28 steps.

Question 5 (2 points)

Note: Make sure to complete Question 2 before working on Question 5, because Question 5 builds upon your answer for Question 2.

Implement the CornersProblem search problem in searchAgents.py. You will need to choose a state

representation that encodes all the information necessary to detect whether all four corners have been reached. Now, your search agent should solve:

python pacman.py -l tinyCorners -p SearchAgent -a fn=bfs,prob=CornersProblem

python pacman.py -l mediumCorners -p SearchAgent -a fn=bfs,prob=CornersProblem

To receive full credit, you need to define an abstract state representation that does not encode irrelevant

information (like the position of ghosts, where extra food is, etc.). In particular, do not use a Pacman GameState as a search state. Your code will be very, very slow if you do (and also wrong) .

Hint: The only parts of the game state you need to reference in your implementation are the starting Pacman position and the location of the four corners .

Our implementation of breadthFirstSearch expands just under 2000 search nodes on mediumCorners . However, heuristics (used with A* search) can reduce the amount of searching required.

Question 6 (3 points)

Note: Make sure to complete Question 4 before working on Question 6, because Question 6 builds upon your answer for Question 4.

Implement a non-trivial, consistent heuristic for the CornersProblem in cornersHeuristic.

python pacman.py -l testSearch -p AStarFoodSearchAgent

Note: AStarFoodSearchAgent is a shortcut for -p SearchAgent -a

fn=astar,prob=FoodSearchProblem,heuristic=foodHeuristic.

You should find that UCS starts to slow down even for the seemingly simple tinySearch. As a reference, our implementation takes 2.5 seconds to find a path of length 27 after expanding 5057 search nodes.

Question 7 (4 points)

Note: Make sure to complete Question 4 before working on Question 7, because Question 7 builds upon your answer for Question 4.

Fill in foodHeuristic in searchAgents.py with a consistent heuristic for the FoodSearchProblem. Try your agent on the trickySearch board:

python pacman.py -l trickySearch -p AStarFoodSearchAgent

Our UCS agent finds the optimal solution in about 13 seconds, exploring over 16,000 nodes.

Any non-trivial non-negative consistent heuristic will receive 1 point. Make sure that your heuristic returns 0 at every goal state and never returns a negative value. Depending on how few nodes your heuristic expands, you'll get additional points:

Remember: If your heuristic is inconsistent, you will receive no credit, so be careful! Can you solve

mediumSearch in a short time? If so, we're either very, very impressed, or your heuristic is inconsistent.

Again, you cannot use the mazeDistance function for this question.

Suboptimal Search

Sometimes, even with A* and a good heuristic, finding the optimal path through all the dots is hard. In these cases, we'd still like to find a reasonably good path, quickly. In this section, you'll write an agent that always greedily eats the closest dot. ClosestDotSearchAgent is implemented for you in searchAgents.py, but it's missing a key function that finds a path to the closest dot.

Question 8 (2 points)

Implement the function findPathToClosestDot in searchAgents.py. Our agent solves this maze (suboptimally!) in under a second with a path cost of 350:

python pacman.py -l bigSearch -p ClosestDotSearchAgent -z .5