Google DeepMind has introduced D4RT, a system that tackles the challenge of teaching AI to perceive the world in four dimensions — three spatial axes plus time. The approach unifies two traditionally separate tasks: reconstructing a dynamic scene's geometry and appearance, and tracking how points in that scene move across frames.
According to the DeepMind blog, D4RT is up to 300 times faster than previous methods for this kind of work. That efficiency could make it practical for applications like robotics, autonomous driving, or augmented reality, where real-time understanding of moving environments is critical.
Because the source is a single announcement, there are no competing claims to compare. The post positions D4RT as a step toward more capable AI perception of dynamic scenes, though it does not provide benchmarks on specific hardware or detailed comparisons beyond the speedup figure.