Research map / Reinforcement learning / Visual and representation RL
GLIB: Efficient Exploration for Relational Model-Based Reinforcement Learning via Goal-Literal Babbling
Rohan Chitnis, Tom Silver, Joshua B. Tenenbaum, Leslie Pack Kaelbling, Tomás Lozano-Pérez
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 GLIB for exploration instead of prior exploration methods | Relational model-based RL · Blocks, Gripper, Keys and Doors, Exploding Blocks, Triangle Tireworld, PyBullet | planning success rate | Improved |
| Use GLIB for exploration instead of prior exploration methods | Relational model-based RL · Six evaluated domains | prediction error | Improved |
| Use lifted goals rather than ground goals | GLIB · All domains except Triangle Tireworld | planning success rate | Improved |
| Use ground goals rather than lifted goals | GLIB · Triangle Tireworld | planning success rate | Improved |
| Use novelty-based goal sampling rather than repeatedly sampling goals without novelty filtering (ablation) | GLIB · Relational model exploration | overall performance | Improved |
| Filter out static and mutex goals (ablation) | GLIB · All evaluated domains | runtime | Improved |
| Filter out static and mutex goals (ablation) | GLIB · All evaluated domains | planning success rate | No clear effect |
| Set the lifted-goal literal bound to k=2 rather than k=1 (ablation) | GLIB lifted · Evaluated domains | planning performance | Improved |
| Use lifted rather than ground novelty in a domain with few predicates (ablation) | GLIB · Triangle Tireworld | planning success rate | Hurt |
| Plan toward babbled goals using the current learned model | GLIB · Keys and Doors | sample efficiency | Improved |
Closest papers by meaning
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Representation-Based Exploration for Language Models: From Test-Time to Post-Training (ICLR 2026 · similarity 0.87)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.87)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.86)
- Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs (NeurIPS 2025 · similarity 0.86)
- Text2Reward: Reward Shaping with Language Models for Reinforcement Learning (ICLR 2024 · 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.
