“Look Out! Behind You…”
Post #3 (04 July 2017): AI Engine Finished…Now the Demo
Jumping right in, we’ve been very busy at Fuzion Labs. The AI engine is completed and working great. However, instead of showing videos of the engine in action, we intend to release a Dystopia demo showing off some of its capabilities. Since the start of this year we have been working diligently to build the Dystopia demo.
The world of Dytopia is making its way from paper to 3D. Check out a few environment screenshots below. These screens are the initial pass of a building that players will visit in the demo. The final version will include environment props, patrolling AI, shaders and additional dirt and grim – of course. More to come this year! We’ll keep you posted!
Post #2 (03 Sept 2016): Sensing the Player’s Presence
Things have been a little slower getting the AI engine completed, as I’ve transitioned to a new job and introduced my 4th child into the world. In addition the team spent a lot of time creating the Dystopia teaser which can be viewed here. All in all, however, the team is satisfied with the progress made and anticipate a new video demonstrating more complete AI sometime this Fall. We are in the process of assembling a vertical slice to demonstrate initial gameplay (planned target date for completion is this Fall). The vertical slice will be used by the team to aide decision making for optimizing Dystopia’s gameplay mechanics (deceive, evade, survive). The vertical slice will allow for experimenting with unique player and AI encounters and validate gameplay mechanics before going into full production.
This latest AI video comes as a result of 5 additional months of work (so from Dec to May) again at a pace of about 5hrs per week. From May to this passed August we’ve also been adding new features to the AI engine, but they won’t be ready for demonstration until later this Fall.
In this video you will observe the AI’s ability to sense the player’s presence. This is achieved through the use of multiple view cones, vision signals, and an alert manager. The AI enters three different awareness levels: unaware, alert, aware/attack. Unaware is represented by having no “eye” sprite above the AI’s head. As the AI starts to become aware of a presence you will observe the “eye” sprite to begin to color-in. Once the AI is aware of a presence, the “eye” changes to yellow which signifies that the AI is alert to a presence but doesn’t know what the presence is. And finally a color of red signifies the AI knows the presence and will attack. Note, however, that this is an alpha so you will observe poor animation blending and some minor bugs.
This video demonstrates the very basic building blocks that will be used to generate complex AI. This will be evident in future blog posts intended for release later this Fall 2016. Enjoy and provide feedback if you have the time!
Post #1: Why Are We Building An AI Engine?
After producing five game prototypes and one officially released mobile game (checkout our games section on our website), we were trying to determine our next move. We learned many lessons from these prototypes and one release which allowed us to identify what we were good at, what our constraints/scope were for projects as a team, and our weaknesses. The next logical step was to conduct a SWOT (strength, weakness, opportunity, threats) analysis to help us identify what our next project should be. In so doing, we came up with the project code named Dystopia. It capitalizes on our team strengths, minimizes our weaknesses and takes advantage of numerous opportunities in the indie game market.
So the AI engine became a requirement for project Dystopia. Since then, I’ve spent an average of 8hrs a week over the last 3 months designing and implementing an AI engine. To chart this course I wrote out 3 simple objectives for the engine:
Must allow AI to sense (e.g. receive sounds, visions) their environment
Must allow AI to interact and affect their environment
Must allow AI to engage the player in unique and exciting ways
These are high level objectives that served as a roadmap for the last three months. I summarize each objective as sensing, action execution, and planning respectively. I’ve managed to tackle the first 2 objectives and lay a foundation for the third.
This first video demonstrates (AI Engine Blog Video One) a baselining of the engine. The baseline provides the basic necessities for any AI to exist in a game. More importantly, the baseline lays the foundation for novel capabilities to be incrementally built into the engine.
One crucial thing I’d like to point out is how the Unity3D engine has allowed us to focus our team resources on innovation. The Unity3D engine provides indie game developers with the ability to focus on designing games and less on implementing core engine technologies at a significantly low cost. For example, Unity’s animation state machine mecanim and pathfinding capabilities. As the sole AI engine developer on the team, with a small number of hours to work, not having to develop a custom animation state machine system and path finding system, allows the team to focus on building novel systems to support innovative game mechanics.
The current baseline of the engine achieves objectives (1) and (2). Our AI now have the ability of hearing and seeing things in their environment and executing actions or interacting with the environment. These objectives were achieved through development of a sensor manager, event manager, and a cognitive architecture that incorporates cognition states and action trees. The integration of these systems with Unity’s pathfinding and mecanim components completed the baseline. The combined effort of these system components are demonstrated in the video.
At the start of the scene illustrated in step one of Fig 1., the AI loads a scripted course of action (CoA). This CoA is determined by the state the AI is currently in and is identified as the best CoA for achieving some goal. The state dictates the goal the AI is trying to achieve. In this case, the AI is in an attack state, and we wanted to simulate a CoA for meleeing the player. The CoA is a set of action trees. Each action tree as executed, will drive Unity’s pathfinding and mecanim system. Yes, our action trees support compound and sequential action execution.
The colorful rays protruding from the AI, Fig 2., are their multiple view cones that model peripheral vision, direct vision, and vertical up and down vision . We have a debugging script that allows us to visualize the cones so we know where they are looking when vision signals are received.
With an AI engine baseline now implemented, we are busily designing and implementing more AI actions (e.g. open doors, room inspections, etc). In the very near future we will release a new video showcasing a player versus AI scenario, where the AI has a larger set of actions at its disposal to engage the player. Stay tuned-in by visiting our AI Engine development blog.