![[2602.14344] Zero-Shot Instruction Following in RL via Structured LTL Representations](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14344] Zero-Shot Instruction Following in RL via Structured LTL Representations
Summary
This article presents a novel approach to zero-shot instruction following in reinforcement learning (RL) using structured linear temporal logic (LTL) representations, enhancing task generalization and performance.
Why It Matters
The research addresses a significant challenge in reinforcement learning, where agents must perform tasks they haven't been explicitly trained on. By leveraging structured LTL representations, this work aims to improve the efficiency and effectiveness of RL agents in complex environments, making it relevant for advancements in AI and robotics.
Key Takeaways
- Introduces a method for zero-shot instruction following in RL.
- Utilizes structured LTL representations to enhance task generalization.
- Implements a hierarchical neural architecture with an attention mechanism.
- Demonstrates superior performance in complex environments.
- Addresses limitations of existing generalist policies in capturing LTL structures.
Computer Science > Machine Learning arXiv:2602.14344 (cs) [Submitted on 15 Feb 2026] Title:Zero-Shot Instruction Following in RL via Structured LTL Representations Authors:Mathias Jackermeier, Mattia Giuri, Jacques Cloete, Alessandro Abate View a PDF of the paper titled Zero-Shot Instruction Following in RL via Structured LTL Representations, by Mathias Jackermeier and 3 other authors View PDF HTML (experimental) Abstract:We study instruction following in multi-task reinforcement learning, where an agent must zero-shot execute novel tasks not seen during training. In this setting, linear temporal logic (LTL) has recently been adopted as a powerful framework for specifying structured, temporally extended tasks. While existing approaches successfully train generalist policies, they often struggle to effectively capture the rich logical and temporal structure inherent in LTL specifications. In this work, we address these concerns with a novel approach to learn structured task representations that facilitate training and generalisation. Our method conditions the policy on sequences of Boolean formulae constructed from a finite automaton of the task. We propose a hierarchical neural architecture to encode the logical structure of these formulae, and introduce an attention mechanism that enables the policy to reason about future subgoals. Experiments in a variety of complex environments demonstrate the strong generalisation capabilities and superior performance of our approach. ...