Robots in the Maze - 9/10

Context

In a research center for autonomous systems, several types of robots are being tested. For these experiments, various types of arenas were used, and each robot follows a specific exploration strategy. During each experiment, the robot is equipped with a set of sensors and monitoring systems that collect information about its activity. The recorded data describes different aspects of the robot's behavior and the environment in which it operates.
These can include numerical values (e.g., average speed, distance traveled, number of collisions, or items collected), but also categorical or descriptive information (such as arena type or environmental conditions). By analyzing this data, patterns can be observed in how robots move, explore space, collect items, or react to obstacles.

Analyzing this data is important for:

• identifying the strategy followed by a robot based on observed behavior;
• understanding the features that differentiate strategies from each other;
• building models capable of automatically recognizing exploration patterns.

To solve this problem, you will have two files at your disposal:

• train.csv – contains labeled examples used for analysis and model training;
• test.csv – contains the same variables, without the strategy_label column, and is used for generating predictions.

Robot Strategies

Four types of strategies are used in the experiments:

• explorer – a robot that tries to visit as many different areas as possible;
• collector – a robot that focuses on collecting items;
• guardian – a robot that patrols and frequently returns to the same regions;
• sprinter – a robot that moves quickly, emphasizing speed and coverage in a short time.

Dataset Description

Each row in the train.csv and test.csv files corresponds to an experiment.

The dataset contains columns describing different aspects of the robot's behavior. Examples of such features are:

Requirements

Task 1 - Number of Arena Types (15 points)

Using the data from the train.csv file, determine how many distinct values appear in the arena_type column.

Task 2 - Maximum Observed Speed (15 points)

Using the data from the train.csv file, determine the maximum value in the avg_speed_mps column.

Task 3 - Most Frequently Used Arena (10 points)

Based on the data from the train.csv file, determine which arena type appears most frequently in the arena_type column.

Task 4 - Maximum Collection Performance (10 points)

Based on the data from the train.csv file, determine the maximum value of the items_collected column.

Task 5 – Identifying the Robot Strategy (50 points)

Build a model that determines the robot's strategy for each experiment in test.csv.

Possible values are:

1explorer2collector3guardian4sprinter

Submission File Format

The submission file must be in csv format and contain the following columns:

Possible values for subtaskID are:

For Task 1 – Task 4, the result is a value calculated globally based on the data in train.csv.
In the submission file, this must be sent only once, using the special value GLOBAL for robot_id.

For Task 5, the predicted strategy for each robot in test.csv must be reported.

Possible values are:

1explorer2collector3guardian4sprinter

Example submission.csv file

1robot_id,subtaskID,answer2GLOBAL,1,103GLOBAL,2,4.314GLOBAL,3,forest5GLOBAL,4,76TR_1000,5,explorer7TR_1001,5,guardian8TR_1002,5,collector

Evaluation

The total score is 100 points.

Task 1 – Task 4 (50 points)

For the first four tasks, the answers are determined exclusively based on the data in train.csv.

Evaluation is done by exact comparison (equality) between the value submitted in the file and the correct value calculated from train.csv.

Task 5 – Identifying the Robot Strategy (50 points)

For this task, the correctness of the predicted strategies for the robots in test.csv is evaluated.

The Macro-F1 metric is used, which calculates the F1 score for each class and then averages these values.

For a class c:

The final Macro-F1 score is the average of the F1 scores for all classes:

where K is the number of classes. In this problem, K = 4.

The score awarded for Task 5 is:

1score = 50 x MacroF1 (if MacroF1 < 0.93)2score = 50 (if MacroF1 >= 0.93)

⬇ Download Starter Kit