Hindsight Experience Replay

E98482

experience replay method reinforcement learning technique

Hindsight Experience Replay is a reinforcement learning technique that improves sample efficiency by reinterpreting failed attempts as successful experiences toward alternative goals.

Jump to: Statements Referenced by

Statements (45)

Predicate	Object
instanceOf	experience replay method ⓘ reinforcement learning technique ⓘ
abbreviation	HER NERFINISHED ⓘ
aimsTo	enable learning from sparse rewards ⓘ improve sample efficiency ⓘ reuse failed trajectories as successful ones for alternative goals ⓘ
appliedTo	multi-goal environments ⓘ robotic manipulation tasks ⓘ sparse reward environments ⓘ
assumes	goals can be derived from achieved states ⓘ
category	off-policy data augmentation technique ⓘ
citationCountCategory	highly cited reinforcement learning method ⓘ
compatibleWith	Deep Deterministic Policy Gradient NERFINISHED ⓘ Deep Q-Learning NERFINISHED ⓘ actor-critic methods ⓘ
coreIdea	reinterpret failed attempts as successful experiences toward different goals ⓘ
field	machine learning ⓘ reinforcement learning ⓘ
implementedIn	OpenAI Baselines NERFINISHED ⓘ Stable Baselines NERFINISHED ⓘ
improves	data efficiency of reinforcement learning agents ⓘ learning speed in sparse reward settings ⓘ
influenced	Goal-Conditioned HER variants ⓘ Hindsight Policy Gradients NERFINISHED ⓘ multi-goal RL benchmarks such as Fetch environments ⓘ
introducedInPaper	Hindsight Experience Replay NERFINISHED ⓘ
keyMechanism	relabelling goals in stored trajectories ⓘ
modifies	replay buffer sampling strategy ⓘ
operatesOn	goal-conditioned policies ⓘ
proposedBy	Alex Ray NERFINISHED ⓘ Bob McGrew NERFINISHED ⓘ Filip Wolski NERFINISHED ⓘ Jonas Schneider NERFINISHED ⓘ Josh Tobin NERFINISHED ⓘ Marcin Andrychowicz NERFINISHED ⓘ OpenAI researchers ⓘ Peter Welinder NERFINISHED ⓘ Rachel Fong NERFINISHED ⓘ
publicationYear	2017 ⓘ
publishedAtConference	NeurIPS 2017 NERFINISHED ⓘ
relatedTo	Universal Value Function Approximators NERFINISHED ⓘ goal-conditioned reinforcement learning ⓘ
requires	goal representation in state space ⓘ
uses	experience replay buffer ⓘ off-policy reinforcement learning ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

OpenAI Baselines → implementsAlgorithm → Hindsight Experience Replay ⓘ