Research map / Reinforcement learning / Visual and representation RL
EAGER: Asking and Answering Questions for Automatic Reward Shaping in Language-guided RL
Thomas Carta, Pierre-Yves Oudeyer, Olivier Sigaud, sylvain lamprier
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 EAGER question-generation and question-answering reward shaping | PPO · Unlock-Medium | average reward: 0.8903 vs 0.8765 | Improved |
| Use EAGER question-generation and question-answering reward shaping | PPO · PutNextTo and Sequence tasks | performance | Improved |
| Use EAGER question-generation and question-answering reward shaping | PPO · Unlock tasks | performance | Improved |
| Combine EAGER reward shaping with RIDE | EAGER · Unlock-Medium | average reward | Improved |
| Combine EAGER reward shaping with RIDE | EAGER · PutNext-local and PutNext-medium | average reward | Hurt |
| Train QA on a wide distribution of successful trajectories rather than a narrow distribution (ablation) | EAGER · BabyAI tasks | sample efficiency | Improved |
| Add a no_answer response to the QA (ablation) | EAGER · PutNextTo | sample efficiency | Improved |
| Reward answers by confidence instead of using a binary reward (ablation) | EAGER · PutNextTo | sample efficiency | Improved |
| Remove neutralization of intrinsic reward on successful trajectories (ablation) | EAGER · PutNextTo-Local | final performance | Hurt |
| Use QA trained on unseen rather than seen goals | EAGER QA · Sequence-Medium | QA success rate: 0.74 -> 0.67 | Hurt |
| Use the QA training dataset to train an agent with behavioural cloning | Behavioural cloning · PutNextTo-Local | success rate: 0 | Hurt |
| Use the QA training dataset with IQL | IQL · PutNextTo-Local | success rate: 0 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RIDE — RIDE: Rewarding Impact-Driven Exploration for Procedurally-Generated Environments (ICLR 2020) · better on 1 of 1 shared result
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
- Offline RL by Reward-Weighted Fine-Tuning for Conversation Optimization (NeurIPS 2025 · similarity 0.87)
- MA-RLHF: Reinforcement Learning from Human Feedback with Macro Actions (ICLR 2025 · similarity 0.87)
- Dense Reward for Free in Reinforcement Learning from Human Feedback (ICML 2024 · similarity 0.86)
- ToolRL: Reward is All Tool Learning Needs (NeurIPS 2025 · similarity 0.86)
- Progress Reward Model for Reinforcement Learning via Large Language Models (NeurIPS 2025 · similarity 0.86)
- Enhancing Question Generation through Diversity-Seeking Reinforcement Learning with Bilevel Policy Decomposition (AAAI 2025 · similarity 0.86)
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.
