Meta-World

Meta-World is a benchmark that contains a variety of tasks used to evaluate multi-task reinforcement learning algorithms, particularly in sparse-reward settings.

Papers using Meta-World (8)

Centralized Reward Agent for Knowledge Sharing and Transfer in Multi-Task Reinforcement Learning2024 · 1 citesLearning Bilevel Policies over Symbolic World Models for Long-Horizon Planning2026Data-Efficient Multitask DAgger2025Goal-Oriented Skill Abstraction for Offline Multi-Task Reinforcement Learning2025Continual World: A Robotic Benchmark For Continual Reinforcement Learning2021 · 20 citesProcedural Generalization by Planning with Self-Supervised World Models2021 · 4 citesTask-Agnostic Continual Reinforcement Learning: Gaining Insights and Overcoming Challenges2022 · 2 citesHierarchical Transformers are Efficient Meta-Reinforcement Learners2024