Robotic Data Collection
Overview
Robotic data collection studies how to obtain useful observations and actions for training and evaluating robots in settings such as manipulation, navigation, and human-robot interaction. Common approaches include teleoperation, wearable and egocentric human capture, autonomous collection guided by policy failures, and simulation or synthetic-data generation. Datasets increasingly combine video, robot state, force, and tactile signals, while quality checks and benchmarks test whether trajectories are feasible and useful on real hardware. The central tension is scaling data efficiently without losing the physical, sensor, and embodiment alignment needed for reliable transfer.
Latest papers
178 papers28 papers in the last four weeks, up 12% on the four weeks before. 0.2% of all new papers.