Meta-World

Meta-World is a benchmark dataset that contains a variety of simulated robotic manipulation tasks used to evaluate reinforcement learning algorithms and reward design methods.

Papers using Meta-World (3)

LaGO: Latent Action Guidance for Online Reinforcement Learning2026When are LLMs Sufficient Policy Optimizers for Sequential RL Tasks?2026A Large Language Model-Driven Reward Design Framework via Dynamic Feedback for Reinforcement Learning2024