A new method called Chronos Policy allows robots to learn actions by directly considering physical time, moving beyond traditional approaches that implicitly embed time within fixed-rate sequences. Developed by researchers at Stanford University, this system addresses a key limitation in robot control: the assumption that observations and actions occur at regular intervals. This assumption breaks down when sensor data rates fluctuate, communication delays vary, or processing times change, leading to potential errors in a robot's understanding of its environment and its own actions. The Chronos Policy makes physical time explicit by conditioning its learning process on the actual capture times of sensor observations and the desired execution times of actions. This is achieved by integrating time-aware rotary positional encodings (time-aware RoPE) into diffusion transformers, a type of neural network architecture. The researchers explored two methods for incorporating time: adding timestamp embeddings to token features or encoding time within the positional encoding itself. They found that time-aware RoPE proved more effective, allowing the robot's attention mechanisms to directly consider physical time without altering the content of the observed data. During training, the Chronos Policy employs temporal randomization. This means that the system is exposed to a wider range of sensing and execution schedules by randomly varying when observations are sampled and when action sequences begin. This training method enables the policy to learn from asynchronous and multi-rate data, eliminating the need to force all sensor inputs onto a rigid temporal grid. Experiments conducted in both simulation and real-world scenarios demonstrated Chronos Policy's effectiveness. The system showed improved performance on dynamic tasks, where precise timing between observations and actions is critical, and on multisensory tasks, where the relative timing of different sensor inputs is important. For standard, quasi-static tasks, the policy maintained its performance while exhibiting greater resilience to changes in sensing and inference timing. The Chronos Policy's ability to handle temporal variations is particularly relevant for robots operating in complex, real-world environments where perfect synchronization is rare. By grounding robot policies in physical time, this research offers a path toward more adaptable and reliable robotic systems capable of performing tasks with greater accuracy and robustness.