AWhat changed
A person can demonstrate a robot-arm movement by moving a smartphone while watching a video stream. RoboTurk connects that familiar input to cloud-hosted robot simulations through low-latency communication. The 2018 platform collects a pilot set of more than 2,200 successful demonstrations and studies whether those examples help reinforcement learning explore sparse-reward tasks. Its contribution is the teleoperation and collection pipeline with its documented learning use, separate from the earlier ideas of imitation learning, WebRTC and PPO.
Scientific scope
A person can demonstrate a robot-arm movement by moving a smartphone while watching a video stream. RoboTurk connects that familiar input to cloud-hosted robot simulations through low-latency communication. The 2018 platform collects a pilot set of more than 2,200 successful demonstrations and studies whether those examples help reinforcement learning explore sparse-reward tasks. Its contribution is the teleoperation and collection pipeline with its documented learning use, separate from the earlier ideas of imitation learning, WebRTC and PPO. The 2018 contribution uses simulated robots. A later real-robot extension has different authors and results and is not included in this award. The interface comparison involves eight students; failing to detect a difference from a VR controller does not prove equivalence. Cross-Pacific trials take longer. The paper gives inconsistent 20- versus 22-hour platform totals, so no reconciled exact total is asserted here. Demonstrations come from human workers; aggregate trajectory hours are not elapsed time. Learning returns are not success percentages, and benefits are not monotonic for every demonstration count.
Evidence
The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. The 2018 contribution uses simulated robots. A later real-robot extension has different authors and results and is not included in this award. The interface comparison involves eight students; failing to detect a difference from a VR controller does not prove equivalence. Cross-Pacific trials take longer. The paper gives inconsistent 20- versus 22-hour platform totals, so no reconciled exact total is asserted here. Demonstrations come from human workers; aggregate trajectory hours are not elapsed time. Learning returns are not success percentages, and benefits are not monotonic for every demonstration count.
Contributors
- The named authors in the research record · SCIENCE: Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.
Provisional WRS assessment, edition 2026-10-23.6. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction