Research map / Reinforcement learning / Visual and representation RL
Instructing Goal-Conditioned Reinforcement Learning Agents with Temporal Logic Objectives
Wenjie Qiu, Wensen Mao, He Zhu
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 GCRL-LTL instead of LTL2Action on in-distribution partially-ordered tasks | goal-conditioned agent · LetterWorld | success rate: 0.661 -> 0.996 | Improved |
| Increase LTL formula depth beyond training distribution (5 to 15) | GCRL-LTL vs LTL2Action · LetterWorld | success rate: 0.477 -> 0.986 | Improved |
| Increase number of conjuncts (4 to 12) out of distribution | GCRL-LTL vs LTL2Action · LetterWorld | success rate: 0.567 -> 0.992 | Improved |
| Use GCRL-LTL instead of LTL2Action on avoidance tasks | goal-conditioned agent · ZoneEnv | success rate: 0.934 -> 0.958 | No clear effect |
| Avoid one extra zone in second stage (OOD task) | GCRL-LTL vs LTL2Action · ZoneEnv | success rate: 0.342 -> 0.928 | Improved |
| Long sequence of four consecutive zones (OOD task) | GCRL-LTL vs LTL2Action · ZoneEnv | success rate: 0.060 -> 0.926 | Improved |
| Same OOD sequence task, discounted return | GCRL-LTL vs LTL2Action · ZoneEnv | discounted return: 0.020 -> 0.220 | Improved |
| Use a single goal-conditioned agent instead of per-specification DiRL training | GCRL-LTL vs DiRL · Ant-16rooms | training steps: 3e6 total vs 3e6 per specification | Improved |
| Train GCRL-LTL without any LTL task sampler (only random goals) | GCRL-LTL vs LTL2Action · LetterWorld/ZoneEnv multi-task | success rate | Improved |
| Apply to omega-regular loop and stability tasks never seen in training | GCRL-LTL · ZoneEnv | omega occurrences | Improved |
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.
- LTL2Action — LTL2Action: Generalizing LTL Instructions for Multi-Task RL (ICML 2021) · 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
- Zero-Shot Instruction Following in RL via Structured LTL Representations (NeurIPS 2026 · similarity 0.92)
- DeepLTL: Learning to Efficiently Satisfy Complex LTL Specifications for Multi-Task RL (ICLR 2025 · similarity 0.91)
- One Subgoal at a Time: Zero-Shot Generalization to Arbitrary Linear Temporal Logic Requirements in Multi-Task Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Automaton Constrained Q-Learning (NeurIPS 2025 · similarity 0.89)
- Reinforcement Learning with LTL and $\omega$-Regular Objectives via Optimality-Preserving Translation to Average Rewards (NeurIPS 2024 · similarity 0.88)
- RLZero: Direct Policy Inference from Language Without In-Domain Supervision (NeurIPS 2025 · 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.
