3D Backscatter Localization for Fine-Grained Robotics
summary
TurboTrack is a 3D backscatter localization system that combines a Bayesian space-time super-resolution algorithm and a time-resolution localization helper (USRP X310 + USRP N210 and an EPC Gen2 reader) to reduce per-packet phase/time ambiguity from ~10,000 candidates to a tractable set. It achieves median sub-centimeter accuracy on x/y/z, 90th-percentile <2 cm in NLOS, 99th-percentile latency <7.5 ms at 300 Hz, validated over ~5,000 trials and demoed on robotic arm handovers, packaging, and nanodrone tracking.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:28Robots are moving towards fine-grained tasks
- 1:09Today's robotic systems rely on vision for identification and localization
- 2:22Can we use RFID localization for fine- grained robotics tasks?
- 4:26Turbo Track
- 4:54Localization by Estimating Distance
- 5:35Approach 1: Measure Time-of-Flight
- 7:20Approach 2: Measure Phase
- 10:48Bayesian Super-resolution
- 12:17Bayesian Fusion using MoGs
- 13:46Space-Time Super-Resolution
- 16:26Idea: Combine standard RFID reader with localization helper that has higher time resolution
- 17:35Quantitative Evaluation
- 18:113D localization accuracy (Los)
- 18:51Partial Implementations
- 19:30Tracking Error vs Speed
- 20:13Robotic Tasks: Robot Arms
- 20:27Conclusion
related talks
26mFrom Panoramic Vision to Lightweight 3D Worlds for Semantic Visual Tracking
10mEnabling Robot-assisted Motion Capture with Human Scale Tracking Optimization
16m"Putting a New Spin on
32mROVINA
06m3D Point Cloud Generation with Millimeter-Wave Radar
47mCore-sets for Real-Time Tracking using Caratheodory Theorem, with Applications to Drones
10mREGROUP: A Robot-Centric Group Detection and Tracking System
47mAchieving Software Quality Excellence Thru Software-Hardware Co-Design & Co-Verification