AI entrepreneur is developing agents that can plan ahead for the unexpected (www.technologyreview.com)

AI entrepreneur Danijar Hafner is quietly developing a groundbreaking robotics startup focused on enabling machines to plan for and respond to unexpected scenarios. Drawing on his expertise in model-based reinforcement learning, Hafner’s approach involves creating world models that mimic physical realities, allowing robots to navigate untested environments—like homes with unfamiliar layouts. This innovation is significant as it could revolutionize how robots operate in human spaces, minimizing the need for extensive real-world training and enhancing their practical applications. Hafner’s previous successes include projects like Dreamer 2 and Dreamer 4, which showcased the potential of agents trained within simulated environments to achieve human-level performance in complex tasks, such as playing video games and mining in Minecraft. His latest venture aims to extend this technology beyond virtual settings, with robots capable of adapting to real-world conditions autonomously. As Hafner hints at ambitious objectives to effect meaningful change, the AI/ML community is keenly watching how this endeavor will evolve and potentially reshape the interaction between robots and the environments they operate in.