Runway launches research push on general world models
Runway says it is starting a long-term research effort on "general world models" that can simulate a wide range of real-world situations, and is hiring for it, though it has released no model or results.
Runway announced a research effort on what it calls "general world models." These are AI systems that build an internal representation of an environment and use it to simulate future events in it.
The company says world-model research has mostly stayed inside limited, controlled settings. Those include toy simulated worlds such as video games and narrow domains such as driving. Runway wants models that can represent and simulate "a wide range of situations and interactions, like those encountered in the real world."
Runway calls its own video generation systems, such as Gen-2, "very early and limited forms" of general world models. To make realistic short clips, Gen-2 has picked up some understanding of physics and motion. The company also says the model still struggles with complex camera and object motion.
Runway lists several open research problems it is working on:
- generating consistent maps of an environment
- navigating and interacting within those environments
- capturing the dynamics of the world's inhabitants as well as the world itself, which means building realistic models of human behavior
The post is a statement of research direction and a hiring call. Runway says it is "building a team to tackle those challenges" and asks interested researchers to get in touch. It gives no model, benchmark, timeline, team size or budget. So this shows where the company plans to focus, not what it can do today.