Assignment 1 - FIT5222 Planning and automated reasoning
Assignment 1: Flatland Challenge
In this assignment your job is to schedule a set of trains through a railway network.
You need to coordinate every train from its starting station to its destination as quickly as
possible. As there can be many trains moving at the same time you need to guarantee that
each path is collision-free.
The assignment has three sections, in increasing order of difficulty. The amount of points
relative to each question is stated in the question heading. A passing grade is 50%.
Be sure to watch the introductory video on Moodle and to read the Introduction to
Flatland documentation which we have prepared for you. Both are available from
Moodle.
You must update your flatland code and installation before starting the assignment:
● Under flatland folder
● git pull
● python3 setup.py install
● Now, run command python3 -m pip list , you should see flatland-rl version is
updated to 2.2.4
● (Hint, use python instead of python3 if python points to the right one on your
machine)
Instructions to get code base (in assignment1_2023 branch of piglet repo) for
assignment is in the Introduction to Flatland documentation.
When the contest server is ready, you will see your f score upon submission.
QUESTION 1: Warm up (15 points)
You are given start and target locations, one at a time. Your job is to route each train
independently from all the rest. In this question collisions are not possible and there is no
time dimension.
For this question, you need to implement a successor function for the Flatland domain. You
also need to choose an algorithm to help you find paths. You are free to use any of the
techniques we have discussed in the lectures, that you have read about in the literature or
can write your own new approach.
Your solution will be evaluated on 40 evaluation instances (only staff have these instances)
with 2 hours timelimit and each instance has 1 agent. We will compare your sum of
individual cost(SIC, for agents did not arrive goal location, its cost will be
Assignment 1 - FIT5222 Planning and automated reasoning
of the map) and success rate to an optimal solution
= 8 * (? + ??)
implemented by teaching team and calculate your score using following method:
= ( __ +
* _)/_
=
/
where is the p score of staff implementation. Generally, a good solution has small
score and large score.
Your final score will be (since there are 15 points available for this question).
× 15
QUESTION 2: Easy mode (25 points)
You are given start and target locations, one at a time, as well as a set of existing paths for
trains that are already moving. Your job is to route each train individually while avoiding
collisions with all the rest. You are free to use any of the techniques we have discussed in
the lectures, that you have read about in the literature or can write your own new approach.
For this question, you need to modify your successor function to account for time. In
addition, there might be the situation that the search algorithm failed to find a feasible
solution as dynamic obstacles block all possible paths. Just return an empty list in this case.
Furthermore, each action and each location for every computed plan need to be
collision-free.
Your solution will be still evaluated on 56 instances with a 2 hours timelimit and your score in
this question will again be computed as the sum of individual path costs (SIC) and compared
to the best solution from students (and staff)!
Your will be computed in the same way as for Question 1. But there are some
differences:
● Here we compute for each instance.
● Each instance contains multiple agents.
● Your final points will be , where is instance id.
∑
÷ 56 × 25
There are up to 25 points available for this question.
Assignment 1 - FIT5222 Planning and automated reasoning
QUESTION 3: Challenge (60 points)
You are given sets of start and target locations at the same time. Your job is to route all the
trains simultaneously in a way that is collision-free. But, each agent has an expected
arrival time, late arrival will result in a penalty.
You are free to use any of the techniques we have discussed in the lectures, that you have
read about in the literature or can write your own new approach.
Now, as all agents are under your control, you need to make all agents reach their goal
locations.
In this question agents may run into malfunctions during execution. The evaluator will call
the replan function when a new malfunction occurs. Implement the replan function to
properly handle malfunction. Refer to the “Introduction to Flatland” document for details
about malfunction and replan function.
Your solution will be still evaluated on 56 instances with different difficulty levels in 2 hours
and your score in this question will again be computed as the sum of individual path costs
(SIC) and compared to the best solution from students (and staff)!
Your computation is similar to Question 1. But there are some differences:
●
= ( __ + +
* _)/_
● The is . 2 * __
● Each instance contains multiple agents.
● Each instance’s will refer to a baseline implementation and an advanced
implementation (or best student solution which ever is better):
There are up to 60 points available for this question.
Report (50 points)
You need to create a report that describes your approach to each of the questions. This
includes a textual description of your approaches, why you adopted that particular approach
and a thorough discussion along with any supplementary material required (such as pseudo
code, images, graphs, tables…).
Assignment 1 - FIT5222 Planning and automated reasoning
Report Marking Rubric