Research map / Reinforcement learning / Visual and representation RL
LTL2Action: Generalizing LTL Instructions for Multi-Task RL
Pashootan Vaezipoor, Andrew C Li, Rodrigo A Toro Icarte, Sheila A. Mcilraith
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 LTL progression rather than processing the unprogressed instruction with a recurrent policy | LTL2Action · Avoidance Tasks, increased depth | total reward: -0.03 -> 0.98 | Improved |
| Use LTL progression rather than processing the unprogressed instruction with a recurrent policy | LTL2Action · Partially-Ordered Tasks, increased depth | total reward: 0.03 -> 0.97 | Improved |
| Use a GNN formula encoder instead of a GRU formula encoder (ablation) | LTL2Action with progression · Avoidance Tasks, increased depth | total reward: 0.42 -> 0.98 | Improved |
| Use a GNN formula encoder instead of a GRU formula encoder (ablation) | LTL2Action with progression · Partially-Ordered Tasks, increased depth | total reward: 0.29 -> 0.97 | Improved |
| Pretrain the LTL module before downstream training (ablation) | LTL2Action with GNN encoder · LetterWorld | sample efficiency | Improved |
| Encode and optimize the full LTL task rather than selecting only the next proposition myopically | LTL2Action · Avoidance Tasks, IID | total reward: 0.88 -> 0.98 | Mixed |
| Replace one-hot proposition features with fixed low-dimensional vectors (ablation) | LTL2Action · LetterWorld, tasks with unseen letters | mean return: $-0.374 \pm 0.021$ -> $0.667 \pm 0.015$ | Improved |
| Use a GRU rather than a GNN to encode formulas (ablation) | LTL2Action · LetterWorld | wall-clock efficiency: 2-3$\times$ | Improved |
Later papers that compared with this method
2 later accepted papers report LTL2Action as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- DeepLTL: Learning to Efficiently Satisfy Complex LTL Specifications for Multi-Task RL (ICLR 2025 · better on 3 of 4 shared results)
- Instructing Goal-Conditioned Reinforcement Learning Agents with Temporal Logic Objectives (NeurIPS 2023 · better on 12 of 12 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
- DeepLTL: Learning to Efficiently Satisfy Complex LTL Specifications for Multi-Task RL (ICLR 2025 · similarity 0.94)
- Zero-Shot Instruction Following in RL via Structured LTL Representations (NeurIPS 2026 · similarity 0.92)
- One Subgoal at a Time: Zero-Shot Generalization to Arbitrary Linear Temporal Logic Requirements in Multi-Task Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Skill-SD: Skill-Conditioned Self-Distillation for Multi-turn LLM Agents (NeurIPS 2026 · similarity 0.88)
- AREAL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning (NeurIPS 2025 · similarity 0.88)
- Supervised Reinforcement Learning: From Expert Trajectories to Step-wise Reasoning (ICLR 2026 · similarity 0.87)
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.
