R2D3 and Hard-Eight: recurrent replay learning from demonstrations

What changed

Mixed demonstration and self-generated recurrent replay with distributed off-policy Q-learning, and introduced eight sparse-reward partially observed tasks with varied initial conditions.

Scientific scope

Mixed demonstration and self-generated recurrent replay with distributed off-policy Q-learning, and introduced eight sparse-reward partially observed tasks with varied initial conditions. Builds on R2D2 and demonstration learning. Demo/agent replay ratio is sensitive; solving several Hard-Eight tasks is not all-task exploration closure or deployment safety.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Builds on R2D2 and demonstration learning. Demo/agent replay ratio is sensitive; solving several Hard-Eight tasks is not all-task exploration closure or deployment safety.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Making Efficient Use of Demonstrations to Solve Hard Exploration Problems

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction