Psst, Play Aggressively: Visor Launches AI for In-Game Alerts in Overwatch'

Multiplayer first-person shooters can be both exciting and nerve-racking. So, it helps to have a knowledgeable player by your side with suggestions.

That's the promise of Visor, which is offering in-game AI alerts and post-game analysis for titles like Blizzard's Overwatch, a first-person shooter for team battles.

San Francisco-based Visor recently scooped up $4.7 million in funding for its AI that aims to help improve your game in Overwatch. The startup in August released the open beta download for its software that offers gamers pointers on their playing.

Visor recently participated in the winter class of the Y Combinator accelerator program.

The startup was founded by Anhang Zhu and Ivan Zhou. The duo met at the University of California, Berkeley, where they coded a lot of projects together and shared a love of gaming.

Visor is like having a really good friend sit next to you while you play and give you feedback. It's surfacing information so that you can act on it, said Zhou, the startup's chief executive and co-founder.

These in-game alerts arrive as text cues that appear on the right-hand side of the screen. They offer tips such as play more aggressively or use your ult (which gives players more power) to help in the exact moment of the game.

We can see through a secondary predictive algorithm that will predict how likely you are at winning a specific game, said co-founder Zhu, a former Facebook engineer.

It turns out that surfacing gaming insights requires training a deep neural network on a lot of gaming data.

Training AI for Gaming

There's no shortage of user-generated gaming data. People upload boatloads of footage to Twitch, YouTube and other online destinations for watching game play. That's allowed Visor to train its AI on more than 500 million frames of video.

Visor uses the video frames to train convolutional neural networks for image classification and recurrent neural networks for predictions of in-game actions. The company used the k-nearest neighbors algorithm, or k-NN, to help support the pattern recognition for its deep neural networks.

The team of six engineers at Visor trained the system locally on NVIDIA GPUs. The initial process of collecting data and training its models took about six months, and it's all deployed now on AWS. The startup gets millions of frames per day to continue honing the models.

How AI Helps Players

Visor tracks data on wins and loses of players. It can use in-game data to determine a player lost by 5 percent, for example. It can asses that several key areas could be improved in the game to win. Visor uses that data to offer one or two tips in those areas to help improve in the game.

The Visor platform was built to be game agnostic, so its deep neural networks could be applied to other games to give players a boost. The founders say there will definitely be additional game titles for Visor. And they point to the precedence for industry acceptance of such tools in professional games with the third-party add-ons for Hearthstone and World of Warcraft.

Visor's in-game intel for Overwatch is aimed at everyone from newbies to professional gamers.

The user experience is the final frontier for what we are doing, said Zhou. We're trying to make it so that you have more fun playing the game.

LINK:	https://blogs.nvidia.com/blog/2018/09/21/visor-ai-gpus-in-game-alerts-...
	See more stories from nvidia

Most recent headlines