Human-like Avatars Expedite Game Development

How Human-like Avatars came to be

Players and avatars have had a relationship with each other for quite some time now, in fact, Google abbreviates this relationship as PAR, which is the interaction between a real player and a digital entity. As a developer, you know this digital entity is what we call an avatar, who is only alive in the digital world of a game.

Now, for a little trip down memory lane, human-like avatars began as an experiment where developers were looking for ways to extend user engagement by giving them the ability to create an avatar that could look like the players, involving the human player even more.

In fact, back in 2007, a humanoid avatar played a competitive game of table tennis that synchronized its movements to resemble those of a human in a fast-paced simulation game. The avatar had correction algorithms that helped the system predict the movements of the paddle and glasses based on velocities across several frames.

As time progressed, this interaction between humans and digital avatars has evolved, leading developers and Deep Reinforcement Learning experts to further explore what Avatars can achieve and improve in terms of game development.

Human-like Avatars in Game Development

In this context, human-like avatars are created as a simulation to play specific game levels, helping you improve content quality and playtest or balance with superior speed compared to human playtesting or game balancing. In short, avatars are digital representations and prospective substitutes for human engagement in simulated game environments.

These human-like agents produce human-like behavior where skill and style are the key parameters to measure human likeness.

With GameAI™, we’ve created and trained Human-like avatars with Deep Reinforcement Learning for testing, game balancing, level design, automated QA, etc. Avatars can do all of these tasks and more in only a few hours instead of weeks, opposite to traditional methods, helping save up to 20x of time and resources.

For example, in games with tens of levels released each week, Deep Reinforcement Learning-based avatars can help boost content quality significantly by ensuring each level is correctly balanced. How?

Well, typically, human playtesters provide biased feedback on each newly-designed level and the designer takes that feedback cycle to improve the game, which can take weeks to achieve so the game is both enjoyable and challenging for players. With Deep Reinforcement Learning avatars automate playtesting and balancing to ensure each level is accurately balanced according to specific parameters set by the game designer.

Game balancing is one of the most time-consuming tasks for a game designer that is actually more routine than creative. Most of the invested time is deposited in waiting for playtesting feedback results.

Human-like avatars learn to play the game as humans would, but at scale, a much, much bigger scale. Thanks to the speed and scale of running these human-like avatars, game designers have access to feedback in much shorter time frames, reducing weeks-worth of waiting to a matter of minutes.

Here are the tangible benefits of human-like avatars in game development:

• Content quality: Faster playtests allow for more iterations per level, which improves the quality of the content you produce. You have more time to refine a game and do it more quickly thanks to the accurate results from human-like avatars. Automated playtesting and balancing frees up your time so you can focus on creative, strategic tasks rather than routine, time-consuming ones.
• Accurate QA: Game designers can use human-like avatar results to explore levels at scale and find bugs quickly. It also helps ensure newly-released features don’t break other aspects of the game, increasing quality as a whole.
• Comprehensive testing: Human playtesters can only playtest so much, right? Well, human-like avatars can go on and on, with no need to take breaks. This introduces the opportunity to test the game more thoroughly and free of human biases.

Building capable, accurate, fast, reliable human-like avatars is no easy feat, but the results are oh so satisfying. Deep Reinforcement Learning can model several behaviors — newbie, skilled, or expert — to balance technique and skills. As the game goes through multiple iterations, human-like avatars are powerful enough to perform much faster sessions, playing thousands of times more quickly, and allowing the exploration of much more game space.

You can pick and choose which sequence of actions you want the avatar to explore, from passing a specific level, exploring different behaviors, or playing a game from start to finish, it all helps you improve the quality of your game.

GameAI™ trains human-like avatars to adapt to changing environments, helping with data generation for game balance and feature evaluation to improve game development.

After seeing the benefits of a human-like avatar, it’s a no-brainer for many to employ these virtual agents into their game development efforts, often leading to better content and fewer headaches from bug testing or hunting.

To sum up:

• Human-like avatars began as an experiment where developers were looking for ways to extend user engagement by giving them the ability to create an avatar that could look like the players;
• Avatars are digital representations and prospective substitutes for human engagement in simulated game environments;
• Human-like agents produce human-like behavior where skill and style are the key parameters to measure human likeness;
• Game balancing is one of the most time-consuming tasks for a game designer that is actually more routine than creative;
• Human-like avatars learn to play the game as humans would, but at scale, a much, much bigger scale;
• Deep Reinforcement Learning can model several behaviors to balance technique and skills with no bias;
• GameAI trains human-like avatars to adapt to changing environments, helping with data generation for game balance and feature evaluation to improve game development.

For additional resources, please visit our FAQ & list of articles.