This video, Part 2 of Rerun's gentle introduction, focuses on visualizing and querying time-series data, leveraging Rerun’s powerful time-handling capabilities essential for robotics and autonomous systems. The tutorial builds a real-time animation spanning 360 frames, demonstrating how to log dynamic multimodal data, including 2D point positions, colors, sizes, and corresponding scalar values (X/Y coordinates and rotation angle). The demonstration walks through the code structure, showing how to log static vs. dynamic elements, and highlights key features of the Rerun viewer. These features include timeline scrubbing, entity path hierarchies (for structuring data components), and synchronized data views (2D spatial plot, time series charts, and raw data frame inspection), all crucial for understanding how complex data evolves over time.
Visualizing Data over Time: A Gentle Introduction to Rerun (Part 2)
0:00 Introduction to Time Handling: The video focuses on Rerun's time-series logging capabilities, crucial for handling multimodal data in robotics and autonomous systems. The goal is to build an animation of rotating points to demonstrate these features.
0:53 Animation Preview: The completed animation, consisting of 360 frames, is shown, featuring rotating 2D points that change color and size over time. The interactive nature of the viewer is highlighted, including scrubbing the timeline by frames, enabling looping (1:33), and zooming (1:40).
2:04 Code Review and Static Data: The script's structure is reviewed, focusing on the logging aspects. The points are animated over 360 frames, changing color, size, and position. An "origin" point (2:55) is logged as static data, which Rerun stores efficiently as it does not change over time.
3:08 Dynamic Logging: The core loop calculates the points' new locations (rotation in space) and updates their attributes (RGB color, radius) for each of the 360 frames before logging the data at that specific time step.
4:32 Multimodal Logging of Scalar Values: The script introduces logging scalar values (X/Y position components and the rotation angle). Logging these values results in the automatic generation of time series charts within the Rerun viewer.
5:57 Entity Path Hierarchy: The viewer's entity structure is explained, demonstrating how data is organized hierarchically (e.g., position/point_zero/x component), allowing users to drill into specific structured data elements, which is vital when sensors log data sparsely or asynchronously.
7:29 Data Inspection: The video shows how to display and use the Data Frame view, which allows users to inspect the raw logged data (like the four position values) at the selected moment in time, aiding in querying and validating the underlying temporal data.
8:19 Key Takeaway: Visualizing data across multiple synchronized views (2D spatial plots, time series charts, and raw data frames) is fundamental for tasks like debugging robotic perception, validating machine learning models, and understanding autonomous decision-making processes.