/ BLOG
Notatki z terenu o uczeniu robotów.
Sub-100ms Robot Teleoperation: A Latency Engineering Guide
How we keep glass-to-glass teleoperation latency under 100ms across continents: the full budget from capture to command, predictive streaming, edge relays, and jitter buffering.
LeRobot vs RLDS vs ROS Bags: Picking a Robot Data Format
MCAP, LeRobot, RLDS, HDF5 — a practical comparison of robot data formats: random access, streaming, ecosystem fit, and why we bet on open standards over lock-in.
cze 2026 · 7 min czytaniaRESEARCHReproducible Robot Datasets: From Capture to Citation
Provenance, versioning, datasheets, and DOIs — the practices that make a robot dataset survive peer review and reproduce across labs.
maj 2026 · 7 min czytaniaPRODUCTOne Operator, Many Robots: Scaling Fleet Teleoperation
Designing jump-in / hand-back supervision so one human can shepherd a fleet: intervention scheduling, confidence signals, and UIs that scale operators, not headcount.
maj 2026 · 7 min czytaniaGUIDECollecting Contact-Rich Manipulation Data (2026 Guide)
Force, tactile, and the events that make grasps learnable: sensor choices, capture rates, operator technique, and labeling for contact-rich imitation learning.
kwi 2026 · 7 min czytaniaENGINEERINGSensor Time Sync for Robot Learning: PTP, Triggers, Sub-ms
Hardware timestamps, clock drift, PTP vs NTP, camera triggers, and the sync architecture that puts a dozen sensors on one clock — so your dataset isn't lying to your policy.
kwi 2026 · 7 min czytaniaCOMPANYThe Robot Data Bottleneck Is a People Problem
Robotics has data scaling laws too — but the constraint isn't storage or formats. It's skilled human time. Scaling collection means scaling operators, not headcount.
mar 2026 · 5 min czytania