“Genpop’s aspirations to build a AAA quality game with only a handful of engineers necessitates a lean development plan and clever solutions to time consuming problems. We’ve envisioned our Data Pipeline with the idea that we’re not only building a solid foundation for analytics, but also to fuel our development of technologies to let us focus less time on ‘making the game work’ and spend more time on ‘making the game good’.”

Abstract

To fulfill Genpop Interactive’s goal of creating a truly original next-gen shooter, we're leveraging modern ML/AI techniques at all levels of the game from Simulating potential Metas to empowering our QA with on-demand artificial playtesters. All good modeling approaches start with good data. We began building a data pipeline early in development: within a month of our internally playable demo, we started dumping activity logs and graphing basic stats. We’ve come a long way since then.

Real-Time Data Capture

Data capture is a valuable tool for video game development because it allows us to collect and analyze data about how players are interacting with the game in real-time. But how real-time are we talking? Traditional data pipes built on Extract/Transform/Loads refresh nightly, but we’re making a competitive shooter where both success and failure is measured in milliseconds, our data-pipeline should be able to keep up with our game. We’re striving for sub-100ms event to graph latency, a lofty goal but this threshold allows us to track behaviors and be alerted to anomalies long before they become ‘game-breaking’.

What do we do with the Data? - Analysis

Relevant analytics are instrumental in helping us identify which parts of the game are over or under performing relative to expectations. By analyzing what players are experiencing in real-time, we can adjust the key parameters of the game to ensure that it is both engaging and balanced.

1. Improving player experience: Timely data can help us identify issues with the game that are impacting player experience. For example, if players are getting stuck on an errant piece of geometry, developers can use real-time data to identify the specific point where players are struggling and make changes to improve the experience.
2. Tracking player behavior: Real-time data capture can help developers understand how players are interacting with the game. By analyzing data on things like player movements, button presses, and in-game purchases, we can gain insights into player behavior that can be used to improve the game.
3. Feature Development: Captured data from internal tests can be used to build new features in the game before they are released. By collecting data on how playtesters and developers are interacting with new features, we can make adjustments and ensure that they are working as intended before the game is released to the public.

But what about our Robot Overlords?

Genpop’s aspirations to build a AAA quality game with only a handful of engineers necessitates a lean development plan and clever solutions to time consuming problems. We’ve envisioned our Data Pipeline with the idea that we’re not only building a solid foundation for analytics, but also to fuel our development of technologies to let us focus less time on ‘making the game work’ and spend more time on ‘making the game good.

Player Simulation for Load Testing and Map Validations (QA Bots)

We knew that time-old challenges like server-load characterization, behavior-driven-crash prevention, and map exploit identification would be time-consuming, especially for a small team, so we focused on designing solutions to them early in the development process. 

To address these issues and more, we've been developing AI driven bots capable of playing our game as our players do. Nicknamed Simulacra, these bots will be pre-trained with human behaviors we’re collecting from our data-pipeline to then be refined and evolved by using adversarial reinforcement learning techniques

Using these bots, we can provide scalable and cost-effective gameplay samples while quickly identifying potential issues and exploring potential exploits that may not manifest until later in the game's lifecycle. Overall, we believe that Simulacra will help us save time and resources while ensuring that our game meets high-quality standards.

AI to Facilitate Game Balance and Novel Game Modes (Meta Simulations)

In multiplayer shooter games, a stagnant metagame can cause players to become bored and leave the game. Game designers must still drive long-term changes to the metagame through balancing, map changes, and other means. Making these balance changes is challenging To address this, and leaning on our Simulacra bot technology we’re developing a platform to give our designers the ability to explore potential meta-strategies in minutes and seconds, paving the way for a renaissance in dynamic game balancing and prediction. While this tech is in its early stages, we're excited to further develop this one in particular. We’re confident that hypothesis engine technologies like this could revolutionize the way game studios approach game balancing, enabling them to rapidly iterate and test new strategies to keep players engaged and interested.