Designing games with the help of machine learning means better content for players, released more often. See how the Angry Birds Dream Blast team incorporates machine learning into their design process.

Article Inside Rovio video-series 17.05.2019

Angry Birds Dream Blast launched in January, 2019 as a super accessible and instantly fun bubble popping puzzle game. Dream Blast introduced a new art style for the Angry Birds characters, representing each member of the core flock as adorable young versions of themselves.

Through the course of the game, players pop groups of similarly colored bubbles to complete a set of challenges. To match the dreamy theme, the Dream Blast gameplay was designed to provide players with a satisfying and relaxing experience any time they have a free moment. Unlike other matching puzzle games, Dream Blast’s levels are not grid-based. Instead, matchable items flow freely within the confines of the level’s border.

We wanted to create a new experience by adding loose physics and making the board feel more alive and tactile.
Angry Birds Dream Blast Lead Designer, Riku Heino

One similarity that Dream Blast shares with many other puzzle games is that its success relies on providing excellent new content for players on a regular basis. That means a lot of time is spent by the game team creating, playtesting, balancing, and optimizing new levels to be added to the game. To overcome this issue, Rovio Senior Data Engineer, Asko Relas, experimented with deep reinforcement learning to teach bots how to play Dream Blast. They could then compare the bots’ success and failure in Dream Blast levels to those of human players. Next, the bots are “set loose” on new levels to predict how a human player would perform in them.

We want to automate the parts of content production that can be automated and leave the creative parts to the people who are best at it.
Senior Data Engineer, Asko Relas

By training bots to playtest levels instead of humans, the aim is to streamline the production process for new content, allowing the Dream Blast team to put the human touch where it is needed, while letting the bots handle time intensive things like difficulty balancing. This way, the creativity of the team can come through, meaning better content for players, released more often.

Watch the full episode, Inside Rovio – Machine Learning Meets Puzzle Game Design, for the whole story from Senior data engineer, Asko Relas, and Angry Birds dream Blast Lead Designer, Riku Heino.