Anastasis Germanidis joins Sophia Dew to discuss Runway’s world action model and its use of video training to predict robot actions. He argues that learning to predict visual scenesA world model learns how an environment is structured and how it may change in response to actions. can provide useful information about physical tasks.
Anastasis Germanidis describes adaptation across several robot platforms, open-weight research and reported benchmark results. He says fine-tuning is still needed for particular uses and describes a data-efficient adaptation experiment as a company result.
The speakers discuss competition in physical AI and the importance of access to models. Anastasis Germanidis predicts that some general-purpose robot policiesA robot policy is the decision-making function that maps observations and goals to actions a robot should take in its environment. will move into production on simpler tasks, presenting the timeline as his forecast.
Watch on YouTube
