# Gradient Coupling: The Hidden Barrier to Generalization in Agentic Reinforcement Learning

arXiv: https://arxiv.org/abs/2509.23870 · HTML: https://mlsanity.com/paper/2509.23870/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use.
- Builds on 21 of its 25 references that are in the map.
- First posted on arXiv 28 Sep 2025.

## Placed in

- LLM agents and tool use / [Reinforcement learning for LLM agents](https://mlsanity.com/agents/reinforcement-learning-for-llm-agents/)

## Builds on (most cited first, 21 of 21)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning](https://mlsanity.com/paper/2503.09516/) (2025)
- [ALFWorld: Aligning Text and Embodied Environments for Interactive Learning](https://mlsanity.com/paper/2010.03768/) (2020)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [Group-in-Group Policy Optimization for LLM Agent Training](https://mlsanity.com/paper/2505.10978/) (2025)
- [ToolRL: Reward is All Tool Learning Needs](https://mlsanity.com/paper/2504.13958/) (2025)
- [AgentTuning: Enabling Generalized Agent Abilities for LLMs](https://mlsanity.com/paper/2310.12823/) (2023)
- [CodeAgent: Enhancing Code Generation with Tool-Integrated Agent Systems for Real-World Repo-level Coding Challenges](https://mlsanity.com/paper/2401.07339/) (2024)
- [DigiRL: Training In-The-Wild Device-Control Agents with Autonomous Reinforcement Learning](https://mlsanity.com/paper/2406.11896/) (2024)
- [Reinforcement Learning for Long-Horizon Interactive LLM Agents](https://mlsanity.com/paper/2502.01600/) (2025)
- [ReMA: Learning to Meta-think for LLMs with Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/2503.09501/) (2025)
- [RLVMR: Reinforcement Learning with Verifiable Meta-Reasoning Rewards for Robust Long-Horizon Agents](https://mlsanity.com/paper/2507.22844/) (2025)
- [AgentRefine: Enhancing Agent Generalization through Refinement Tuning](https://mlsanity.com/paper/2501.01702/) (2025)
- [MARFT: Multi-Agent Reinforcement Fine-Tuning](https://mlsanity.com/paper/2504.16129/) (2025)
- [Harnessing Uncertainty: Entropy-Modulated Policy Gradients for Long-Horizon LLM Agents](https://mlsanity.com/paper/2509.09265/) (2025)
- [On the Effect of Negative Gradient in Group Relative Deep Reinforcement Optimization](https://mlsanity.com/paper/2505.18830/) (2025)
- [RL Fine-Tuning Heals OOD Forgetting in SFT](https://mlsanity.com/paper/2509.12235/) (2025)
- [Atom-Searcher: Enhancing Agentic Deep Research via Fine-Grained Atomic Thought Reward](https://mlsanity.com/paper/2508.12800/) (2025)
- [Do not Abstain! Identify and Solve the Uncertainty](https://mlsanity.com/paper/2506.00780/) (2025)

## Built on by (most cited first, 3 of 3)

- [HIPIF: Hierarchical Planning and Information Folding for Long-Horizon LLM Agent Learning](https://mlsanity.com/paper/2606.10507/) (2026)
- [Diagnosing Task Insensitivity in Language Agents](https://mlsanity.com/paper/2606.26918/) (2026)
- [Where Do CoT Training Gains Land in LLM based Agents?](https://mlsanity.com/paper/2606.26935/) (2026)

## Abstract (arXiv)

Reinforcement learning (RL) is a dominant paradigm for training autonomous agents, yet these agents often exhibit poor generalization, failing to adapt to scenarios not seen during training. In this work, we identify a fundamental cause of this brittleness, a phenomenon which we term "gradient coupling." We hypothesize that in complex agentic tasks, the high similarity between distinct states leads to destructive interference between gradients. Specifically, a gradient update that reinforces an optimal action in one state can inadvertently increase the likelihood of a suboptimal action in a similar, yet different, state. To solve this, we propose a novel objective where the actor is trained to simultaneously function as a classifier that separates good and bad actions. This auxiliary pressure compels the model to learn disentangled embeddings for positive and negative actions, which mitigates negative gradient interference and improve the generalization performance. Extensive experiments demonstrate the effectiveness of our method.