ParliRobo.github.io

Introduction

Our work focuses on developing participant lightweight AI robots (PARs) for complex real-time gaming. To this end, we collaborate with X-Game, a popular 3D-FPS real-time game that owns ~210,000 users in 2021. Based on comprehensive explorations, we eventually develop a practical PAR system (called ParliRobo) through a novel “transform and polish” methodology, which achieves ultralight implementations of the core system components by non-intuitive yet principled approaches and meanwhile carefully fixes the probable side effect incurred on user perceptions. This repository contains the implementation code of ParliRobo (including the training and deployment code) and our released data.

Implementation of ParliRobo

Currently, we are scrutinizing the codebase to avoid possible anonymity violations. After that, we will release the source code of this study as soon as we have finished examining it and acquired its release permission from the authority. The codebase is organized as follows.

code
|---- math.py
|---- preprocessor.py
|---- training
          |---- KLDiv.py
          |---- categorical.py
          |---- drtrace.py
          |---- entropy.py
          |---- get_shape.py
          |---- mmo.py
          |---- policy_graph.py
|---- deployment
          |---- client.py
          |---- server.py
          |---- GamebotAPI.py
          |---- pressure_test.py
          |---- preprocessor.py

implementation/math.py provides math tools for data clipping and embedding.
implementation/preprocessor.py preprocesses the visual information of participant AI robots by defining functions such as bilinear approximation.
implementation/training contains several modules for training PARs of ParliRobo.
implementation/deployment includes several modules for deploying ParliRobo.

The released part can be found here.

Data Release

Currently, we have released a portion of the representative sample data (with proper anonymization) for reference here. As to the full dataset, we are still in discussion with the authority to what extent can it be released. We will make the rest dataset in public as soon as possible after receiving the permission and desensitizing the dataset.

These data are organized in interaction_delay.xlsx, mini-Turing_Test.xls, and real-time_gaming_information.xlsx, respectively (for detailed data, please click here). For each file, we list the specific information coupled with the regarding description as follows.

sample-dataset
|---- interaction_delay.xlsx
|---- real-time_gaming_information.xlsx
|---- mini-Turing_Test.xls

interaction_delay.xlsx

Column Name	Description
`instruction_id`	Unique ID of each instruction
`system_name`	PAR system name
`interaction_delay`	Average delay of instructions in a second

real-time_gaming_information.xlsx

Column Name	Description
`uid`	Unique ID of each PAR or human player (cannot be related to the user’s true identity)
`solution_name`	PAR systems or human player
`shoots`	Number of shoots performed by each user
`hits`	Number of hits performed by each user
`jumps`	Number of jumps performed by each user
`crouches`	Number of crouches performed by each user
`moving distances`	Length of moving distances of each user
`pros gathered`	Number of pros gathered by each user

mini-Turing_Test.xls

Column Name	Description
`video_id`	Unique ID of each game video clip
`solution_name`	PAR systems or human player
`V1` - `V34`	Unique ID generated to identify a volunteer

Demo Video

We have released a demo video of a participant AI robot (PAR) trained by ParliRobo fighting with a human player. The video was recorded on a mobile device. Note that the protagonist in this video is the PAR.

Platform Requirements

Currently, we build and deploy ParliRobo on Linux servers with proper Python support. Specifically, we leverage TensorFlow, an end-to-end open source platform for machine learning, to train participant AI robots (PARs) in this study.

For Developers

Our code is licensed under Apache 2.0. Please adhere to the corresponding open source policy when applying modifications and commercial uses. Also, some of our code is currently not available but will be released soon once we have obtained the permission.