RoboTurk: A Crowdsourcing Platform for Robotic Skill Learning through Imitation

Source

• Raw Markdown: paper_roboturk-2018.md
• PDF: paper_roboturk-2018.pdf
• Preprint: arXiv 1811.02790
• Project page: roboturk.stanford.edu

Core Claim

RoboTurk is a crowdsourcing platform for collecting 6-DoF teleoperated robot manipulation demonstrations with mobile devices.

Sensor-Time-Series Notes

• The central object is a teleoperated demonstration trajectory with human-generated control inputs over time.
• The paper is useful context for why imitation-learning data in robotics often comes with nonstationary human demonstrations, network-latency effects, and variable task timescales.
• The data interface is closer to action-conditioned trajectory modeling than to passive time-series forecasting.

Links Into The Wiki

• Robotics Time-Series Modeling
• Action-Conditioned Time-Series Datasets

Open Questions

• How should policies distinguish demonstrator style, network latency, and robot dynamics when learning from teleoperation traces?
• Which temporal smoothing or action-chunking methods best reduce compounding error from human demonstration noise?