Creating a game world that contains a realistic simulation of a large human population has been a goal of game design for many years. To make this simulation meaningful to players, it is important that the world can be altered by their actions. Changes must have persistence, so that the effect they have can propagate out to a large area.

Game worlds have been growing in scale, and player expectations of NPC behaviours have been increasing. Traditional approaches to engine architecture limit what can be achieved within tight performance budgets. Limits on NPC numbers often mean that the characters that the player encounters are throw-away copies of a generic type with no concept of their place in the world.

This talk will describe the fundamental technical problems that put limits on the number of fully realised NPCs in our worlds, and describe techniques that can be used to break these limits. We will discuss deterministic approaches, and simulation level of detail as techniques that can be easily applied. We will move on to talk about optimising processing requirements using asynchronous logic processing and highly multithreaded design. We will then touch on the power that can be gained from cloud computing, looking at distributed processing.

Many of the problems of creating believable NPC behaviour have been solved in recent years. By concentrating our effort on increasing the scale of our simulations, we can create worlds that feel truly alive and responsive to the actions of the player.
 

The last few years has seen the term "Artificial intelligence" become increasingly ubiquitous: as sensationalised headlines dominate our media and new projects and initiatives use the term in an effort to garner attention. In amongst all this is video games: an industry that has historically used artificial intelligence as means of performance theatre, with a variety of bespoke tools and methodologies designed to support these goals, but in-turn alienating games from other industries given the unique challenges faced. In the last few years AI has seen major breakthroughs in the area of Machine Learning - and more specifically 'Deep Learning' - but what does it do? How does it work? And more importantly, how is it relevant to the video games we make and how we make them? In this talk we're going to establish the state of play for AI and video games: how it has being used up until now, explaining how machine learning differs from 'classical' AI, the risks and opportunties of deep learning, but also some of the more novel innovations and opportunties emerging in corporate and academic research that will shape our working practices and the games we make in the coming years.

Takeaway:

• An overview of common tools, technologies and applications of AI in the video games industry.
• How we can expect to see deep learning impact the games industry in the next 3-5 years. 
• Tools, technologies and collaborations available to smaller studios in the UK and overseas.

This talk will outline the architecture of the cloud gaming pipeline we developed at Polystream. Attendees will learn why there is a need for an alternative to video streaming and how to virtualise the GPU to stream compressed graphics commands in APIs such as D3D11. We will detail how to stream AAA titles, in under 16 milliseconds, without source changes. A description will be given on how to use Renderdoc for automated graphics testing and how to leverage ISPC to extract the maximum performance out of AVX-512 on cloud servers. We will also discuss the unique challenges and tradeoffs of creating bespoke tools and algorithms for visualizing and compressing real-time GPU data.

Takeaway:

• Detailed technical understanding of the command streaming process and requirements
• Pros and cons of command streaming as a stronger alternative to video streaming for 3D interactive content and applications
• Clarity on cloud architecture for command streaming

The benefits of adding machine learning (ML) to your development arsenal are becoming increasingly hard to ignore. Whether used to adapt gameplay to user tastes and improve retention or to automatically create content, done correctly, its effects on both player and developer can be remarkable. But for studios with no prior ML experience, where should devs begin?

In this talk, Andrew Webb, machine learning engineer at vTime, will take delegates from zero to ML hero with the help of a virtual presenter, created during the session using ML avatar generation.

Offering invaluable advice on getting started, Andrew will take you through what is and isn't achievable with current machine learning tech, data acquisition, and how to make use of freely available and state-of-the-art work for your own projects. The talk will also cover how to get ready for your first ML hire, including preparing the data for initial projects and the hardware and resources you’ll need to hit the ground running.

Takeaway:

• Learn what is and isn‚Äôt possible with current machine learning tech
• Find out how to prepare before your first machine learning hire
• Learn how to make use of state-of-the-art open source ML projects