Research map / Reinforcement learning / MDPs and planning
RUDDER: Return Decomposition for Delayed Rewards
Jose A. Arjona-Medina, Michael Gillhofer, Michael Widrich, Thomas Unterthiner, Johannes Brandstetter, Sepp Hochreiter
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add return decomposition and reward redistribution to PPO | PPO · Bowling | average score: 56 -> 192 | Improved |
| Add return decomposition and reward redistribution to PPO | PPO · Solaris | average score: 616 -> 1,827 | Improved |
| Add return decomposition and reward redistribution to PPO | PPO · Venture | average score: 820 -> 1,350 | Improved |
| Add return decomposition and reward redistribution to PPO | PPO · Seaquest | average score: 1,616 -> 4,770 | Improved |
| Redistribute delayed rewards using return decomposition | RUDDER · Grid World | learning time to 90% of maximal expected return | Improved |
| Redistribute delayed rewards using return decomposition | RUDDER · The Choice | learning time to 90% of maximal expected return | Improved |
| Redistribute delayed rewards using return decomposition | RUDDER · Trace-Back | learning time to 90% of maximal expected return | Improved |
| Use RUDDER instead of the PPO baseline | PPO on Atari · 32 Atari games | learning speed | Improved |
| Use RUDDER instead of the PPO baseline | PPO on Atari · Atari games | average score in the last 10 training games | Mixed |
Later papers that compared with this method
2 later accepted papers report Rudder as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Learning to Identify Critical States for Reinforcement Learning from Videos (ICCV 2023 · better on 1 of 1 shared result)
- EDGE: Explaining Deep Reinforcement Learning Policies (NeurIPS 2021 · better on 7 of 7 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Belief Reward Shaping in Reinforcement Learning (AAAI 2018 · similarity 0.89)
- Off-Policy Reinforcement Learning with Delayed Rewards (ICML 2022 · similarity 0.87)
- Interpretable Reward Redistribution in Reinforcement Learning: A Causal Approach (NeurIPS 2023 · similarity 0.87)
- Efficient Reinforcement Learning in Probabilistic Reward Machines (AAAI 2025 · similarity 0.86)
- Reinforcement Learning with Lookahead Information (NeurIPS 2024 · similarity 0.85)
- The Value of Reward Lookahead in Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
