Research map / Reinforcement learning / Visual and representation RL

Reverse Forward Curriculum Learning for Extreme Sample and Demo Efficiency

ICLR 2024 · Reinforcement learning › Visual and representation RL · approach: Hierarchical skills and priors

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes Reverse Forward Curriculum Learning (RFCL), which uses a few demonstrations to build a reverse-then-forward curriculum for RL on sparse-reward tasks, improving sample and demonstration efficiency.

Problem

RL from sparse rewards needs huge amounts of interaction, and demonstration-augmented RL typically requires many high-quality demonstrations that are hard to obtain.

Approach

Uses a small number of demonstrations to construct a curriculum that starts episodes near the goal state by reversing demonstrations and then progressively moves start states forward toward the initial state, enabling extreme sample and demo efficiency.

Experimental setup

Metrics: sample efficiency demonstration efficiency

Builds on: reverse curriculum generation demo-augmented RL

Closest papers by meaning

• Demonstration-free Autonomous Reinforcement Learning via Implicit and Bidirectional Curriculum (ICML 2023 · similarity 0.89)
• Curriculum Design for Teaching via Demonstrations: Theory and Applications (NeurIPS 2021 · similarity 0.88)
• Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.87)
• Guided Meta-Policy Search (NeurIPS 2019 · similarity 0.87)
• Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.86)
• Environment Design for Inverse Reinforcement Learning (ICML 2024 · similarity 0.86)