Explore a cutting-edge 3D localization system for fine-grained robotic tasks in this USENIX conference talk. Dive into the design and implementation of TurboTrack, which achieves sub-centimeter accuracy in localizing backscatter nodes without constraints on location or mobility. Learn about the system's pipelined architecture that extracts sensing bandwidth from backscatter packets and its Bayesian space-time super-resolution algorithm for accurate positioning. Discover how TurboTrack achieves a median accuracy of sub-centimeter in x/y/z dimensions with low latency, enabling real-time applications in collaborative robotics. Gain insights into practical demonstrations involving robotic arms and nanodrones for indoor tracking, packaging, assembly, and handover tasks.
3D Backscatter Localization for Fine-Grained Robotics
Overview
Syllabus
Intro
Robots are moving towards fine-grained tasks
Today's robotic systems rely on vision for identification and localization
Can we use RFID localization for fine- grained robotics tasks?
Turbo Track
Localization by Estimating Distance
Approach 1: Measure Time-of-Flight
Approach 2: Measure Phase
Bayesian Super-resolution
Bayesian Fusion using MoGs
Space-Time Super-Resolution
Idea: Combine standard RFID reader with localization helper that has higher time resolution
Quantitative Evaluation
3D localization accuracy (Los)
Partial Implementations
Tracking Error vs Speed
Robotic Tasks: Robot Arms
Conclusion
Taught by
USENIX
