Danijar Hafner's stealth startup develops AI agents that plan ahead

danijar hafner, a former google deepmind researcher, has launched a stealth startup aimed at developing ai agents capable of navigating environments they haven't encountered during training. hafner's approach utilizes model-based reinforcement learning, where agents learn within world models that emulate physical reality. these agents can then make predictions about future outcomes, enabling them to handle unfamiliar situations in real-life scenarios. hafner's work has already shown promising results, with his agents achieving human-level performance in atari 2600 games and solving the minecraft diamond challenge. this technology could be crucial for deploying robots in human spaces, such as homes, where they may encounter new and unpredictable situations.