Research map / Reinforcement learning / Imitation learning
PsiPhi-Learning: Reinforcement Learning with Demonstrations using Successor Features and Inverse Temporal Difference Learning
Angelos Filos, Clare Lyle, Yarin Gal, Sergey Levine, Natasha Jaques, Gregory Farquhar
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 |
|---|---|---|---|
| Use ITD to infer rewards from reward-free demonstrations | DQN on Roundabout · Roundabout | normalized return: 0.85±0.02 -> 0.92±0.01 | Improved |
| Use ITD to infer rewards from reward-free demonstrations | DQN on CoinGrid · CoinGrid | normalized return: 0.73±0.02 -> 0.77±0.03 | Improved |
| Use ITD to infer rewards from reward-free demonstrations | PPO on FruitBot · FruitBot | normalized return: 0.37±0.02 -> 0.35±0.04 | Hurt |
| Use ΨΦ-learning for 0-shot task transfer | CoinGrid · R-G and -R+G | return: 0.0±0.0 -> 0.2±0.1 | Improved |
| Use ΨΦ-learning for 0-shot task transfer | CoinGrid · -R-G | return: -1.0±0.0 -> -0.4±0.2 | Improved |
| Adapt using one episode of new-task experience | ΨΦ-learning on CoinGrid · R-G, -R+G, and -R-G | return: 0.0±0.0 -> 1.0±0.0; -1.0±0.0 -> 1.0±0.0 | Improved |
| Add an L1 sparsity penalty on preferences (ablation) | ITD · Offline multi-task IRL | overall performance | No clear effect |
| Use RL interaction to update shared cumulants alongside ITD (ablation) | ΨΦ-learning imitation model · Roundabout held-out demonstrations | action prediction accuracy: little over 80% -> ≈95% | Improved |
| Use demonstrations that are irrelevant or misleading rather than only ego-task data | ΨΦ-learning · Highway adversarial task | performance | No clear effect |
| Combine multi-task demonstrations with online RL and GPI | ΨΦ-learning · Highway multi-task | performance | Improved |
Closest papers by meaning
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.90)
- Non-Adversarial Inverse Reinforcement Learning via Successor Feature Matching (ICLR 2025 · similarity 0.90)
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.89)
- Rethinking Inverse Reinforcement Learning: from Data Alignment to Task Alignment (NeurIPS 2024 · similarity 0.89)
- Dual RL: Unification and New Methods for Reinforcement and Imitation Learning (ICLR 2024 · similarity 0.88)
- Recent Advancements in Inverse Reinforcement Learning (AAAI 2024 · similarity 0.88)
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.
