ChuckeRearmed
Regular
Aside not watching them, I have yet to see them in games though.Why are you wondering when they've had technical deep dives and HotChips presentations about it? Was your memory erased?
Aside not watching them, I have yet to see them in games though.Why are you wondering when they've had technical deep dives and HotChips presentations about it? Was your memory erased?
Aside not watching them, I have yet to see them in games though.
FIFA 22 uses some type of trained machine learning animation system that they're calling "Hyper-motion". It's available on Series X/S and PS5.
https://www.ea.com/games/fifa/fifa-22
That's an entirely different matter than what you posted. The support is there.
Are they doing real-time inference or did they use ML offline to generate the animations? I couldn't tell from that presentation.https://ubm-twvideo01.s3.amazonaws....sentations/Zinno_Fabio_FromMotionMatching.pdf
I suppose they use motion synthesis.
The Series consoles have Schrödinger's ML.
Are they doing real-time inference or did they use ML offline to generate the animations? I couldn't tell from that presentation.
We present a real-time character control mechanism using a novel neural network architecture called a Phase-Functioned Neural Network. In this network structure, the weights are computed via a cyclic function which uses the phase as an input. Along with the phase, our system takes as input user controls, the previous state of the character, the geometry of the scene, and automatically produces high quality motions that achieve the desired user control. The entire network is trained in an end-to-end fashion on a large dataset composed of locomotion such as walking, running, jumping, and climbing movements fitted into virtual environments. Our system can therefore automatically produce motions where the character adapts to different geometric environments such as walking and running over rough terrain, climbing over large rocks, jumping over obstacles, and crouching under low ceilings. Our network architecture produces higher quality results than time-series autoregressive models such as LSTMs as it deals explicitly with the latent variable of motion relating to the phase. Once trained, our system is also extremely fast and compact, requiring only milliseconds of execution time and a few megabytes of memory, even when trained on gigabytes of motion data. Our work is most appropriate for controlling characters in interactive scenes such as computer games and virtual reality systems.
https://arxiv.org/pdf/2111.06449.pdfExpert Human-Level Driving in Gran Turismo Sport Using Deep Reinforcement Learning with Image-based Representation
Abstract
When humans play virtual racing games, they use visual environmental information on the game screen to understand the rules within the environments. In contrast, a state-of-the-art realistic racing game AI agent that outperforms human players does not use image-based environmental information but the compact and precise measurements provided by the environment. In this paper, a vision-based control algorithm is proposed and compared with human player performances under the same conditions in realistic racing scenarios using Gran Turismo Sport (GTS), which is known as a high-fidelity realistic racing simulator. In the proposed method, the environmental information that constitutes part of the observations in conven- tional state-of-the-art methods is replaced with feature representations extracted from game screen images. We demonstrate that the proposed method performs expert human-level vehicle control under high-speed driving scenarios even with game screen images as high-dimensional inputs. Additionally, it outperforms the built-in AI in GTS in a time trial task, and its score places it among the top 10% approximately 28,000 human players.
Given the novelty of what Drivatar represents, it required a unique insight into making it happen. Hence, while Forza Motorsport was developed at Turn 10 Studios, in Redmond, Washington in the United States, the original Drivatar was conceived and developed at Microsoft Research Cambridge, in the United Kingdom. This isn’t the only time that Microsoft Research has had a significant impact on Xbox game development, with the likes of the Kinect peripheral and the TruSkill matchmaking systems built on the AI research conducted by their divisions.
The original Drivatar is quite interesting, given that it exploits a feature unique to the Xbox at that time: a hard drive. Unlike Nintendo’s GameCube and Sony’s Playstation 2, the original Xbox came with an 8GB hard disk installed and ready to go. An idea that became more commonplace with the next generation of consoles. So in the early Forza Motorsport titles 1 through 4, your Drivatar was trained on the hard drive of your console, meaning it only existed on your Xbox. Meanwhile, other drivers are based on existing pre-trained data from the developers at Turn 10 that was shipped on the disc. But as we’ll see in a moment, each Drivatar is fundamentally a really small amount of data. And while they wouldn’t be shared via Xbox Live, you could grab an Xbox Memory Unit and copy the Drivatar for use on other consoles.
My drivatars in fh5 trying to kill me driving into me :dI think it's fantastic that GT is doing something along these lines now. MS started using ML to create AI for Forza Motorsport on the original Xbox.
The Inner Workings of Forza’s Drivatar | AI and Games
While primitive by today's standards, it's interesting in that the training for the AI happened on the Xbox itself.
Forza Motorsport 5 saw a massive upgrade in the Drivatar system as now the AI was trained on Microsoft's servers using Deep Learning with the data inputs for it being every person that played Forza Motorsport while connected to the internet. However, those inputs are also modified by data inputs using data input from actual professional drivers. So they became like a hybrid of you and your friend's drive skill combined with a professional driving skillset.
It's very interesting in how they attempted to create an AI that mimicked you and your friends but attempted to mitigate the especially bad driving habits of players (like deliberately crashing into another player) by blending it with professional data sets.
Of course, the system of creating driver AI using Deep Learning is constantly evolving at Turn 10 so it sees improvements with each Forza release.
Regardless, it's very cool to see Gran Turismo also doing this now.
The X360 generation is when MS really started to leverage ML and Deep Learning for console gaming. It's basically the only reason that Kinect even worked as well as it did and why it represented such a breakthrough in consumer level full body tracking.
Regards,
SB
My drivatars in fh5 trying to kill me driving into me :d
Outperforming stock Gran Turismo AI is nothing to brag about. lol I've seen budget racing games with better/more competitive AI than in GT.
Top 10% of human players is a reasonably ok result if true, but still far from competitive. When more dedicated sims programmed by quite small teams are capable of *way* better AI behavior and performance(both are important), I feel like even just as a tech experiment, this doesn't tell me much since it's an area Polyphony have never seemingly cared much about, so it's kind of a poor basis of comparison in terms of what its real potential is.
I have no doubt ML could ultimately be used to create decent driving AI, but can it do better than the current best human programmers?
the main feature of this approach that ai use same data as human observing track, few years ago they just want to use ai to drive as fast as possible and they made times impossible even for world finalistsOutperforming stock Gran Turismo AI is nothing to brag about. lol I've seen budget racing games with better/more competitive AI than in GT.
Top 10% of human players is a reasonably ok result if true, but still far from competitive. When more dedicated sims programmed by quite small teams are capable of *way* better AI behavior and performance(both are important), I feel like even just as a tech experiment, this doesn't tell me much since it's an area Polyphony have never seemingly cared much about, so it's kind of a poor basis of comparison in terms of what its real potential is.
I have no doubt ML could ultimately be used to create decent driving AI, but can it do better than the current best human programmers?
Drivatars straight up didn't work. I'd love for Microsoft to really open up on what's really going on, but I know for a fact that this 'training' doesn't truly copy your driving technique or anything like that. I think all it must do is feature a sort of *very* rough behavioral system with limited, pre-defined categories(all built-in to the normal programmed AI), and then depending on where on a sort of 'chart' that your Drivatar is figured to be, is how it would apply the AI to a given car.I think it's fantastic that GT is doing something along these lines now. MS started using ML to create AI for Forza Motorsport on the original Xbox.
Ah ok, that's very different to what the article above was claiming. And obviously a lot more impressive. I mean, I've seen programmed AI that is basically 'world class' sort of pace as well, but the ability to do this with fairly simple training is promising for the future. Of course, doing such a thing for a single car/track and only for maximum laptime isn't nearly enough, but it's a good start.It's not stock AI. It's outperforming human expert player and it's not done on ps5.
I notice it rides over the kerb at 0:13It's not stock AI. It's outperforming human expert player and it's not done on ps5.
I notice it rides over the kerb at 0:13
It surprises me, as you would think AI would be be better than say lewis hamilton at driving a car,
eg
this turn I turn 0.1 degrees more to the left 0.001 seconds quicker
does that improve my speed?
this x a billion different permutations to see the best way to handle that 5 seconds of track
Game development presentations - a useful referenceDrivatars straight up didn't work. I'd love for Microsoft to really open up on what's really going on, but I know for a fact that this 'training' doesn't truly copy your driving technique or anything like that. I think all it must do is feature a sort of *very* rough behavioral system with limited, pre-defined categories(all built-in to the normal programmed AI), and then depending on where on a sort of 'chart' that your Drivatar is figured to be, is how it would apply the AI to a given car.