Implicit Curriculum in Procgen Made Explicit
Zhenxiong Tan, Kaixin Wang, Xinchao Wang
Abstract
Procedurally generated environments such as Procgen Benchmark provide a testbed for evaluating the agent’s ability to robustly learn a relevant skill, by situating the agent in ever-changing levels. The diverse levels associated with varying contexts are naturally connected to curriculum learning. Existing works mainly focus on arranging the levels to explicitly form a curriculum. In this work, we take a close look at the learning process itself under the multi-level training in Procgen. Interestingly, the learning process exhibits a gradual shift from easy contexts to hard contexts, suggesting an implicit curriculum in multi-level training. Our analysis is made possible through C-Procgen , a benchmark we build upon Procgen that enables explicit control of the contexts. We believe our findings will foster a deeper understanding of learning in diverse contexts, and our benchmark will benefit future research in curriculum reinforcement learning.
Ask about this paper
Your agent reads all of it.
Lune indexed this paper to the last equation, along with the top-tier papers that cite it. Ask a question and the answer quotes them.
Your agent calls
Luneget_paper_fulltext
Free to start. No credit card required.
Terminal
Install the CLIlune papers fulltext f5a9bde3-5ea3-417c-b601-8a4f45c73532Builds on9
- Leveraging Procedural Generation to Benchmark Reinforcement LearningKarl Cobbe, Christopher Hesse, Jacob Hilton, John SchulmanICML 2020 · 685 citations
- Emergent Complexity and Zero-shot Transfer via Unsupervised Environment DesignMichael Dennis, Natasha Jaques, Eugene Vinitsky, Alexandre M. Bayen et al.NeurIPS 2020 · 362 citations
- Prioritized Level ReplayMinqi Jiang, Edward Grefenstette, Tim RocktäschelICML 2021 · 211 citations
- Evolving Curricula with Regret-Based Environment DesignJack Parker-Holder, Minqi Jiang, Michael Dennis, Mikayel Samvelyan et al.ICML 2022 · 175 citations
- Enhanced POET: Open-ended Reinforcement Learning through Unbounded Invention of Learning Challenges and their SolutionsRui Wang, Joel Lehman, Aditya Rawal, Jiale Zhi et al.ICML 2020 · 148 citations
Related papers
- Adaptive Procedural Task Generation for Hard-Exploration ProblemsKuan Fang, Yuke Zhu, Silvio Savarese, Li Fei-FeiICLR 2021 · 36 citations
- TeachMyAgent: a Benchmark for Automatic Curriculum Learning in Deep RLClément Romac, Rémy Portelas, Katja Hofmann, Pierre-Yves OudeyerICML 2021 · 28 citations
- Curriculum Reinforcement Learning via Constrained Optimal TransportPascal Klink, Haoyi Yang, Carlo D'Eramo, Jan Peters et al.ICML 2022 · 44 citations
- CurBench: Curriculum Learning BenchmarkYuwei Zhou, Zirui Pan, Xin Wang, Hong Chen et al.ICML 2024 · 11 citations
- Diffusion-based Curriculum Reinforcement LearningErdi Sayar, Giovanni Iacca, Ozgur S. Oguz, Alois KnollNeurIPS 2024 · 12 citations
