Star Citizen: AI Roundtable

Each week, Star Citizen Live hosts one or more developers to talk live about the progress of their work. This Friday, the AI ​​team is in the spotlight with no less than five developers on the set. AI work is ahead of other teams, in the sense that most of the technology already exists and only needs to be adapted to a specific gameplay, or even include specific animations. However, there is still work and action to be taken. Here’s what we picked up.

Assignment

Star Citizen’s NPC works through AI assignments. An entity (Server, player, etc.) Defines the work done before running AI. There are situations where AI will enter an infinite loop of assignment in a single container (object, panel, wall, etc. It is a container of the interaction entity like an electric panel with a fuse). This loop is intended to popularize the universe and bring it to life, so AI will then have access to all the actions available on this panel and will implement them one by one to avoid excessive redundancy. AI will be able to communicate with each fuse for example. Moreover, AI is aware of the material it carries, He can then use a flashlight, a tablet or anything useful in a given situation.

NPC crew operation

Recruitment itself is a matter of gameplay and therefore not suitable for this team. Developers, on the other hand, talk about the AI ​​crew’s internal workings:

  • AI will turn a dialogue into an assignment
  • The tasks assigned may be highly specific, or conversely, may be generic depending on the task.
  • When allocating, Players must respect AI’s profession under punishment. A pilot will eventually be able to repair a ship, but with less efficiency than an engineer. If an engineer is given the responsibility of the family, his morale may be damaged etc.

As mentioned in the introduction, the technology is ready. Using a game database is therefore trivial for AI developers, but if gameplay is missing for the player, it is also missing for NPC. Problems must be overcome At the mission or animation level used in case of an error, the “fallback” chain must be precise and effective.

AI environment dependent: Animation must constantly adapt to metrics, it must be able to notice and use the tools available within the range, otherwise it must know where to find them. The animations are so varied and innumerable that it takes time to record and implement.

Dynamic Navmesh plane for the planet

The team has completed the development of a dynamic nebula for the planet. This means that potential paths are constantly calculated around NPCs, and so we will be able to find many more outside the outposts for example in the future. So technically an NPC will be able to transport its ship from one place to another, without it being completely lost or frozen.

Dynamic newbies have already been passed to start working with the design team

AI reinforcement

With the advent of navigation links and dynamic navigation, it will be seen AI reinforcement ships landing at outposts. For now, it’s just a letter, but they’re working on strengthening the ship’s vehicle. Navigation links are currently not for ladders or elevators, only for ramps.

When a hostile AI comes to the site, it is already in battle mode, no waiting period or searching for the enemy. He will be situationally aware (NPC will not try to shoot a ship with a gun), But in the same situation there will be different reactions depending on the “personality” of the NPC (Some will be very aggressive, others are completely new and not experienced in handling weapons, etc.).

Healing by AIs

They are working with AI’s ability to heal friendly units (player or NPC). Care awakens an unconscious character, or worries the release of a prisoner.

Loot by AIs

AI is currently omniscient. So he is able to search for a body for more ammunition. The AI ​​resource management system already exists, so it is not to cross a level, this behavior is connected to the rest of the system. However, developers want to create the illusion of AI by searching for hardware.

Performance

In a continuous universe, AI is completely tied to servers, so it is not uncommon for their performance to decline. If server-mashing brings significant benefits to this level, they must think of solutions to improve performance in:

  • Decreased entity density across locations (the more AI entities in a location, the worse the performance)
  • Fallback relevant social animation in case of very long server response time.
  • Limit the number of counts (requests, questions, etc.) for each image.

Different

  • They want to bring the universe to life, so NPC will have a mood. For example, we see an NPC expressing its frustration for missing its metro.
  • They work on AI animals.
  • The blade is working, but the gameplay is not ready.

Leave a Comment