Mercuna AI Navigation

Creators of the AI Navigation middleware for Games. Mercuna 3D for flying and underwater AI. Mercuna Ground for characters, vehicle and animal AI.

[Mercuna AI Navigation @MercunaDev]

Our new automatic jump generation will identify locations where agents can jump across gaps or down from heights. This drastically increases where your AI can go, and saves designers huge amounts of time no longer needing to place navlinks throughout maps

[Mercuna AI Navigation @MercunaDev]

We are seeking an experienced Unreal Engine/Gameplay Programmer to support our customers using Mercuna in their games and to improve our Unreal integration. If you are passionate about advancing AI navigation software and working with AAA game studios, we want to hear from you! https://t.co/64ILUmQ5hZ

[Mercuna AI Navigation @MercunaDev]

Congratulations to The @ChineseRoom on the release of Still Wakes the Deep! We were delighted they selected Mercuna to drive their monster movement.

[The Chinese Room @ChineseRoom]

🌊 OUT NOW 🌊

December 1975. Disaster strikes an oil rig off the coast of Scotland. Save your crew from an otherworldly horror.

"Still Wakes the Deep" marks The Chinese Room's return to Narrative Horror.

Play @SWTD_Game on PC / consoles now: https://t.co/TjXuKqm3S7

[Mercuna AI Navigation @MercunaDev]

Great to see Dune Awakening showcased in #StateOfUnreal! At Mercuna we’ve been working with Funcom on the navigation challenges for this game for years now.

It’s an awesome IP to be working with!

[Mercuna AI Navigation @MercunaDev]

Cool to see interesting modes of locomotion in Rise of Hydra. #StateOfUnreal
Mercuna can automatically add jump links, allowing NPCs to navigate between rooftops without scripting.

[Mercuna AI Navigation @MercunaDev]

#StateofUnreal is that a lot of developers would like to evaluate its state. Hopefully I get in!

[Mercuna AI Navigation @MercunaDev]

We also have sensible defaults for clamping the position (e.g. clamping further down than up), but this is a hard problem and we’ve certainly had various bugs come up in this area!

[Mercuna AI Navigation @MercunaDev]

Daniel Brewer points out just how tricky looking up the correct poly on the nav mesh can be in his #AISummit micro-talk.
Mercuna uses a quad tree that follows terrain reasonably closely and accelerates the lookup into the coarse nav mesh used for path finding. #GDC2024

[Mercuna AI Navigation @MercunaDev]

Will Chambers has been extending hierarchical goal planners to handle actions that execute in parallel and over time. I now want to give Ara a go to watch the different AI leaders play!
#GDC #AISummit

[Ben Sunshine-Hill @Sneftel]

@MercunaDev I think people are too eager to stuff AI-related data into nav structures. There’s a few areas where that fits, but I think it’s often just because it’s the structure that particular dev is familiar with and controls.

[Mercuna AI Navigation @MercunaDev]

The LLMs are back again at the #AISummit - by adding a per agent memory store, believable agents can be made, and in this case play Werewolf.

The results are interesting, but human interaction didn’t work quite as well as it first appeared to - allowing agents to more often be persuaded by human arguments would be the next step.

[Mercuna AI Navigation @MercunaDev]

Hidden Door presented some impressive results from using a mix of generative techniques, breaking down the task of generating a playable story and then applying appropriate techniques to each part. Both LLMs and Postgres databases can co-exist! #AISummit #GDC2024

[Mercuna AI Navigation @MercunaDev]

Lots of great stuff in this talk, but very custom for the specific game.

I wonder if some generic, designer accessible system could be used to add annotations to hierarchical areas, allowing some more flexible queries in a more generic way.

[Mercuna AI Navigation @MercunaDev]

Impressive work for Suicide Squad, lots of custom systems augmenting Unreal’s capabilities. Another case of customising the Recast flow to generate additional navigation data structures - that’s becoming something of a theme at the #GDC #AISummit.

[Mercuna AI Navigation @MercunaDev]

First up at the #GDC #AISummit this morning, utility systems. A clear explanation of how scoring decisions works and common pitfalls. While decision making isn’t something we normally tackle at Mercuna, this is similar to how we decide on steering options with our context steering implementation.

Definitely some food for thought here!

[Mercuna AI Navigation @MercunaDev]

Last talk of the day at the #GDC #AISummit was about a game that aims to make the experience of training neutral network a core part of the gameplay. It sounds like a fun way to learn about neural network training!

That’s Little Learning Machines if you want to give it a go.

[Mercuna AI Navigation @MercunaDev]

It would be interesting to see whether this really worked in practice - would designers that were less familiar with behaviour tree actually be able to use this to create detailed behaviours, and would the language model handle the complexity as the number of available nodes increases.

[Mercuna AI Navigation @MercunaDev]

More LLMs at the #GDC #AISummit! This time describing desired behaviour in natural(ish) language and having the LLM generate a behaviour tree, and even implementation of the missing behaviour tree nodes.

[Mercuna AI Navigation @MercunaDev]

It would have been interesting to hear more about how the query results were evaluated and compared - this comes down to utility system evaluation which we should hear more about tomorrow morning.

[Mercuna AI Navigation @MercunaDev]

Interesting #GDC #AISummit talk about building modular environment evaluation from Sam Fay-Hunt. The idea is that designers can write queries that plug into multiple different systems.

By moving the execution scheduling of these queries out of the designer’s control they can achieve timeslicing in a generic way.

[Mercuna AI Navigation @MercunaDev]

The solution that Tomorrow Falls settled on relies on offline generation to determine which higher level paths are viable, allowing paths up to 8km long.

Mercuna’s navigation is based on similar principles, but allows annotation of the nav grid with friction, slope and modifier markup. Our maximum path length is 1-2km, but the grid can be regenerated quickly at runtime.

[Mercuna AI Navigation @MercunaDev]

Good to hear of someone else tackling vehicle pathfinding at the #GDC #AISummit. Our solution also allows vehicles to navigate considering their maximum turning rates and also minimising sharp corners to carry maximum speed.