Resources & Blog
Technical guides, research insights, and product updates from the Radioception team.
Why Imitation Learning Needs Better Data: The Annotation Quality Gap
Most off-the-shelf motion datasets were built for computer vision tasks, not for robot manipulation. Here's why annotation fidelity matters for imitation learning — and what to look for.
RLDS TFRecords Format Guide: Structuring Motion Data for Robot Training
A comprehensive walkthrough of RLDS TFRecords structure for motion datasets, including time synchronization, custom message types, and best practices for multi-sensor recording.
Force-Torque Data in Manipulation Tasks: When and Why You Need It
Many manipulation tasks require contact-rich interactions that visual data alone can't capture. We explain when force-torque sensing is critical and how to incorporate it in your training pipeline.
Grasping Policy Generalization: The Role of Subject Diversity in Training Data
Training on demonstrations from a single demonstrator leads to brittle policies. We analyze how demographic and skill-level diversity in training data improves generalization.
Radioception JSON Format v2.3 — What's New
Version 2.3 of our native JSON format adds temporal confidence fields, world-frame coordinate support, and improved hand landmark naming aligned with MediaPipe conventions.
Building a Motion Capture Lab on a Budget: What Robotics Teams Actually Need
Not every team can afford a Vicon system. We break down which capture equipment delivers the most annotation quality per dollar for robot training data specifically.
Ready to get your dataset?
Submit a skill request and receive a custom quote within 24 hours.
Request Custom Dataset →