RuneScape Bench: MMOs for Agents [video]

(youtube.com)

102 points | by evakhoury 1 day ago

15 comments

  • pokpokpok 1 hour ago
    Thanks HN Author here to answer questions if helpful!

    If this inspired you, would encourage people to try it out by pointing their agents to this repo and giving a goal:

    https://github.com/MaxBittker/rs-sdk

    You can also view the realtime activity of the server here: https://maxbittker.github.io/rs-world/

  • Confiks 1 hour ago
    What a great talk. My introduction to programming as a kid was also writing Runescape bots, using some Delphi tool called SCAR [1] (together with writing plugins for my Counter Strike 1.6 server using AMX Mod X). Start at Varrock bank, wizard hut, dialogue picker, teleport, find out where you are, mine, mine, mine, teleport back, back to bank, deposit, and again.

    My great essence mining bot contraption horribly broke when Runescape started randomizing all colors slightly, after making a paltry amount in dollars selling the essences.

    [1] https://scar-divi.com/

    • camdenclark 16 minutes ago
      Yep I got started with programming with SCAR and on srl-forums with SRL as well. Was really fun.
      • kylecazar 11 minutes ago
        Same, but on Kaitnieks/Sythe. Really fond memories of those communities ~2004-2007.
        • camdenclark 10 minutes ago
          I don't think anyone has written online about what these communities were like and how interesting they were. I should do it!
  • almostlit 13 minutes ago
    This is a great talk. I've been working on a very similar concept for World of Warcraft at https://github.com/agent-wow/agent-wow.

    A lot of the techniques and pitfalls he talks about also aligns with my experience on WoW agents too. It's also interesting to see how agents coordinate in these kinds of game environments.

  • ecshafer 34 minutes ago
    This is a good talk. There is some similar things in the private server / self hosted server space for World of Warcraft. There are bots that will play the game, talk with you, etc. using classical game aI, but people are also hooking them up with LLMS for more planning, strategy, and conversation.
    • almostlit 9 minutes ago
      > but people are also hooking them up with LLMS for more planning, strategy, and conversation.

      I did this at https://github.com/agent-wow/agent-wow but haven't yet gotten to the level that the rs-sdk author has with benchmarks and running economy experiments.

      That said, a lot of the key points he talks about in the video align pretty well with my experience too.

  • zshn 1 hour ago
    Was fun exploring this months ago. Back then even the smallest amount of game knowledge you could provide a model would drastically improve their performance. Most of it is in their training data but they hallucinate a ton of incorrect information.
  • infogulch 37 minutes ago
    Games are a huge untapped training potential. Especially sandbox games that have distant goals with a complex dependency tree and structured/semi-open solution space like Factorio. There are so many dimensions that factor in to make a solution good.

    Somebody is working on this for Factorio at: https://www.youtube.com/@FactoryArenaAI

    • hsuduebc2 3 minutes ago
      lol, this person owns a gold mine that he is using Astra? Subscribed.
  • jparishy 3 hours ago
    I've wondered if these types of casual open world, resource collection games-- which are basically a series of mini-games-- would make good model training envs. Each mini-game has a pretty simple mechanic that needs to be learned, as well as the navigation and interactions to begin it. Those could be pretty useful generic skills I think.

    One of my side projects is a game, and I've made it so agents can spawn off their own world shard to play in. There's a little Chrome-driven agent harness that can grab low res screenshots while it speaks directly to the game over the wire to control things.

    I've had Luna play it and it can drive around (you play as a bike) but it did struggle to play one of the mini-games (Disc Golf). My idea was that model trainers could run it an RL loop using the harness to get feedback data and screenshots to use, then reset the game world as needed during the process.

    A smart model capable of multi-modal inference on screenshots/game context at 30-60fps could be pretty interesting, and I'm sure it's coming. No idea how much it would generalize but would be fun to see done.

    • FireCrack 2 hours ago
      I've tried, casually, quite a bit of this. With generally poor results. Often this ends up similar what I have seen as classic RL problems with seeking reward in sparse environments. Unless you give the agent an a-priori goal outside the world it tends to not do anything interesting. Theoretically there ought to be ways around this such as herarchical approaches, but those tend to scale in complexity faster than my limited resources can keep up.

      It's very neat though.

      Edit: I really should watch the video before commenting. Looks like it goes into lots of these problems itself, and they got much further than me.

      • nomel 33 minutes ago
        I think you would have to go back to more classic genetic algorithm/pet training methods: Iterate through increasing "distance" to the goal. Initially make the goal simple/immediately available, then slowly "distance" them from the goal, with more obstacles/etc introduced between, as you go.

        For a homing pigeon, this would be literal, driving further and further each time you release them.

        For an agent that you train to walk, this would be flat surface to past some line, then introduce speed reward, then non-flat surfaces, then obstacles, then curves, etc [1].

        [1] "AI Warehouse" channel is all about training AI for light gaming tasks: https://www.youtube.com/watch?v=L_4BPjLBF4E

  • pprotas 3 hours ago
    Saw it this morning, this video is a goldmine of ideas on how to build agent harnesses and benchmark them.
  • akersten 3 hours ago
    The author did a great job capturing the essence of what made RS such a great game, and the soul that Jagex has totally lost over the years with updates to make the game a softer place.

    This formative moment could never happen today, because the wilderness is behind 3 confirmation screens instead of just being something you get to discover in the game:

    > You have to know that [items] are worth something in order to not get taken advantage of. And it goes to kind of extreme levels too because not only will people, try to give you a bad price, but the game is just really well known for the scamming and the lying and stealing your password. And it's really the wild west out there.

    > And so, and really this is true. My very first memory of the very first time that I entered the world of RuneScape, I was asking somebody for help. I said like I was trying to make bread and I was like, "Where can I find a mill so I can make some bread?" And a guy was like, "Follow me. I'm going to show you." And I followed him like into these woods. And you know, you can imagine he turned around. He started to attack me. And I realized what was happening. I was probably 11 years old. I realized what was happening. And I made a break for it. I ran back south for cover and I got away. I didn't die. And I remember like, you know, closing the game.

    > I think that my parents hadn't been home, so I had time to think about this. And I was like, that was the coolest thing that's ever happened to me. And I think that this is kind of the beauty and the magic of this game is that it's a little bit of like a rough and tumble world where you have to look out for yourself. And as a kid, it's like that's actually really great practice to be out in the world with real people and have to understand what's going on and learn things like, oh yeah, if somebody offers you a really great deal, maybe be careful.

    Cheers to this kind of game design. Wish we'd see it again some day.

    • Eridrus 10 minutes ago
      Truly, the wilderness was the best part of the game.

      I remember grinding away to a set of black armor, going right to the wilderness and getting immediately crushed and losing it immediately lol.

    • Fellshard 2 hours ago
      Similar experience in a MUD. I fled a faction that experienced a hostile takeover, and was subsequently consistently hunted down by one of the leaders of the faction. Rather than it killing my motivation, it forced me to learn the survival skills the game gave to discover and monitor the locations of others, identify incoming presences through various shortcuts, etc. In essence, it made the game more alive to me and forced me to develop my character's story to go with it. Just a totally unique experience.
    • mitxela 2 hours ago
      RS3 became an enshittified hellscape, that's why everyone is on OSRS. Without OSRS the company was enshittifying to its own death.

      The modern game - either one - has more safeguards to make sure you don't enter a dangerous area by mistake. There isn't "three confirmation screens" but there is a ditch you have to click to jump over, and when you do you get the one screen that basically says "hey you know players can attack you past this ditch right? [Acknowledge] [Cancel] [Acknowledge and don't ask again]"

      (Technically it's not even correct because the first few tiles past the ditch are also still in the safe zone)

      There's plenty of enticing stuff in the Wilderness to make you want to go there and accept the danger, you don't really need to be tricking 11 year olds to have them face danger. Though I suppose you're arguing it's better to teach the 11yo that they can be tricked like this, in a video game instead of real life. I agree with that. At the same time it's a pretty unfair game if it has no indication you're walking into a PVP area other than scenery objects nobody pays attention to, and I think they current system is adequate.

      • akersten 2 hours ago
        > in a video game instead of real life.

        Yeah, that's the argument. I'm kind of getting at something deeper though, like when you say this:

        > At the same time it's a pretty unfair game if it has no indication you're walking into a PVP area other than scenery objects nobody pays attention to

        Unfair by a comparison to contemporary game design practices maybe, but my point is that it was more fun before (and paying attention to scenery is a valuable lesson to learn as well).

        Anyway the point isn't that I think Scamming is Good, Actually. The point is that Modern Game Design is Boring, and bringing back some of the discovered elements of gameplay rather than having everything explained in a dialog box or bright yellow safety paint would serve the industry well.

        • mitxela 1 hour ago
          > rather than... bright yellow safety paint

          Ah, yes. The "Pac-Man reimagined as a modern AAA game" problem: https://youtu.be/tVY_oJlv_Eo

          I don't think Runescape goes overboard with warnings. A lot of places have a "hey this place is dangerous, are you sure you want to go in?" dialog but they usually don't even explain exactly what the danger is, they don't stop you from going in unprepared if you click yes, and there were some changes to reduce how much stuff you lose if you die but dying in the game is still overall a huge PITA, especially in PVP where your killer gets to keep most of your stuff. The most impactful softening I can think of is that you can pay a large fine to the Grim Reaper to get your stuff back if you die to monsters and don't get back to claim your gravestone in time.

    • lirolero 2 hours ago
      [dead]
  • zatkin 3 hours ago
    RuneScape was what led me to become a software engineer (through the private server communities) and it's refreshing to see someone describe this game in layman terms, all the way up through to having AI agents drive the player interactions. I worry, though, about Jagex's stance on all of this. Perhaps this is what pushed them towards the recent RuneScape 4 announcement?
    • pageandrew 3 hours ago
      > RuneScape was what led me to become a software engineer (through the private server communities)

      I have the exact same story. Mopar, RuneServer, then SoulSplit (look at what the founder of SoulSplit is doing now...)

      I wonder how many people have the same background?

      • byteknight 3 hours ago
        No sooner than I replied to OP, your message mirrors my experience. I really am interested now how many people had their introduction from similar things.
        • radihuq 2 hours ago
          I’m in a similar boat - made my first private server when I was 11 and loved tech since

          Back in those days I didn’t know about VPSes so I’d host the private server on our family computer. Going to bed I’d keep the computer on and pray that the server stayed on for at least 3 hours before crashing. Worst feeling was waking up and checking the logs to see the server crashed 10 mins after you went to bed :’)

        • loufe 2 hours ago
          It certainly was my first taste of programming and stoked a lifelong interest in computers which has served me very well.
    • Fergusonb 3 hours ago
      The software doesn't work on the live game - it's designed for a "private server" / fan reconstruction based around the 2004 version of the game.

      Jagex almost certainly knows of lost city, but it's not really comparable to OSRS these days. People are likely to play 2004scape and then move to the live version of the game once the nostalgia wears off.

      Jagex's lawyers have been absolutely decimating botting communities in the last 6 months, if they wanted Lost City gone, it would be gone already.

      I wouldn't surprised to see them reach out for the valuable data he's collecting on the SDK's main server.

    • giancarlostoro 3 hours ago
      There were two paths in the 2000s some people went with Runescape, others went with Habbo Hotel. The amount of insanity that was done to Habbo since its inception to date, is astounding. To the point that because the Habbo community kept tinkering with Shockwave, Sulake made Habbo Origins from the fruit of their labor.
    • byteknight 3 hours ago
      I too got started because of RS. Writing scripts for SCAR to automate RS :)
      • camdenclark 7 minutes ago
        This and Simba with SRL. Was a really awesome learning experience and totally formative to becoming an engineer.
      • openair18 2 hours ago
        Stands for “sh* compared to Autorune”. I had autorune from Kaitnieks, that was a cool tool as well.
        • byteknight 2 hours ago
          Kait actually made SCAR too :)

          Only use of Delphi I have ever seen personally.

    • tmpz22 3 hours ago
      Jagex is an interesting finance story if you look into how they're passed around between private equity as one of the UKs largest game studios.

      As a product the inside scoop is that the recent success of the game is in large part due to the community itself. The community development of Runelite, the OSRS wiki, the 117 HD plugin, and YouTube personalities.

      I don't have high hopes that the status quo between Private Equity and the community will sustain itself.

      • gbear605 2 hours ago
        Of note, just today Jagex released a plugin interface for Runescape 3 which was developed together with developers from the Runelite team and the existing Runescape 3 third-party plugin tool (Alt-1). The relationship seems pretty friendly right now.
      • lanyard-textile 3 hours ago
        Given that Jagex (and whoever owns them now) can pull the rug and kill Runelite at any time, it's remarkable there's been some sense of harmony at all. Concessions have been made both ways.

        The relationship isn't perfect. But the fact that there's a relationship and a two way dialogue at all? Unheard of.

  • reactordev 1 hour ago
    If this was for UO, you'd have a bloodbath. RuneScape's systems were pretty tame compared to UO:SA where anywhere outside of towns was PvP allowed and when you died you lost all your gear.
    • ASpring 1 hour ago
      RuneScape took a lot of lessons from UO. Originally though they did allow player-killing anywhere outside towns if you had the pk setting on. This didn't last very long though and they introduced the wilderness idea the game has now.
  • Aeroi 2 hours ago
    full circle. i remember downloading an autominer back in like 2004 to auto click through mining coal in falador.
  • mitxela 2 hours ago
    It's really shocking that there aren't more games like (old-school) RuneScape. Like where is OSRS but you're a mafia boss in NYC, or OSRS but you're on a space station? Obviously games with these themes exist, but not games in this genre, which I'd say is defined by fairly shallow but very broad content building on a cartoonishly simple engine. Modern games seem to be going relatively long on engine capabilities but short on content instead. You'd think AI would have no problem designing 50 good quests for an NYC mob boss in a few days or less.

    Maybe RS's origins from the text MUD era led it to a weird path-dependent place. The other day I watched a whole video essay about how weird it is that everything has an "examine" option when no other graphical game has this, but every MUD has it.

    • ianstormtaylor 38 minutes ago
      > You'd think AI would have no problem designing 50 good quests for an NYC mob boss in a few days or less.

      Why would you think that?

  • gzer0 2 hours ago
    [dead]
  • vibewright 3 hours ago
    [flagged]