When Are RL Hyperparameters Benign? A Study in Offline Goal-Conditioned RL

• URL: http://arxiv.org/abs/2602.05459v1
• Date: Thu, 05 Feb 2026 09:08:17 GMT
• Title: When Are RL Hyperparameters Benign? A Study in Offline Goal-Conditioned RL
• Authors: Jan Malte Töpperwien, Aditya Mohan, Marius Lindauer,
• Abstract summary: We observe robustness to changes in hyper parameter configurations in offline Deep Reinforcement Learning (RL)<n>We find that high sensitivity to changes in hyper parameter configurations during training is not inevitable in RL, but is amplified by the dynamics of bootstrapping.<n>These results suggest that high sensitivity to changes in hyper parameter configurations during training is not inevitable in RL, but is amplified by the dynamics of bootstrapping.
• Score: 9.430246534202857
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Hyperparameter sensitivity in Deep Reinforcement Learning (RL) is often accepted as unavoidable. However, it remains unclear whether it is intrinsic to the RL problem or exacerbated by specific training mechanisms. We investigate this question in offline goal-conditioned RL, where data distributions are fixed, and non-stationarity can be explicitly controlled via scheduled shifts in data quality. Additionally, we study varying data qualities under both stationary and non-stationary regimes, and cover two representative algorithms: HIQL (bootstrapped TD-learning) and QRL (quasimetric representation learning). Overall, we observe substantially greater robustness to changes in hyperparameter configurations than commonly reported for online RL, even under controlled non-stationarity. Once modest expert data is present ($\approx$ 20\%), QRL maintains broad, stable near-optimal regions, while HIQL exhibits sharp optima that drift significantly across training phases. To explain this divergence, we introduce an inter-goal gradient alignment diagnostic. We find that bootstrapped objectives exhibit stronger destructive gradient interference, which coincides directly with hyperparameter sensitivity. These results suggest that high sensitivity to changes in hyperparameter configurations during training is not inevitable in RL, but is amplified by the dynamics of bootstrapping, offering a pathway toward more robust algorithmic objective design.

Related papers

• Controllable Exploration in Hybrid-Policy RLVR for Multi-Modal Reasoning [88.42566960813438]
  CalibRL is a hybrid-policy RLVR framework that supports controllable exploration with expert guidance.<n>CalibRL increases policy entropy in a guided manner and clarifies the target distribution.<n>Experiments across eight benchmarks, including both in-domain and out-of-domain settings, demonstrate consistent improvements.
  arXiv  Detail & Related papers  (2026-02-22T07:23:36Z)
• Consolidation or Adaptation? PRISM: Disentangling SFT and RL Data via Gradient Concentration [56.074760766965085]
  PRISM achieves a dynamics-aware framework that arbitrates data based on its degree of cognitive conflict with the model's existing knowledge.<n>Our findings suggest that disentangling data based on internal optimization regimes is crucial for scalable and robust agent alignment.
  arXiv  Detail & Related papers  (2026-01-12T05:43:20Z)
• The Path Not Taken: RLVR Provably Learns Off the Principals [85.41043469428365]
  We show that sparsity is a surface artifact of a model-conditioned optimization bias.<n>We mechanistically explain these dynamics with a Three-Gate Theory.<n>We provide a parameter-level characterization of RLVR's learning dynamics.
  arXiv  Detail & Related papers  (2025-11-11T18:49:45Z)
• The Choice of Divergence: A Neglected Key to Mitigating Diversity Collapse in Reinforcement Learning with Verifiable Reward [57.56453588632619]
  A central paradox in fine-tuning Large Language Models (LLMs) with Reinforcement Learning with Verifiable Reward (RLVR) is the frequent degradation of multi-attempt performance.<n>This is often accompanied by catastrophic forgetting, where models lose previously acquired skills.<n>We argue that standard RLVR objectives lack a crucial mechanism for knowledge retention.
  arXiv  Detail & Related papers  (2025-09-09T06:34:32Z)
• Adaptive $Q$-Network: On-the-fly Target Selection for Deep Reinforcement Learning [18.579378919155864]
  We propose Adaptive $Q$Network (AdaQN) to take into account the non-stationarity of the optimization procedure without requiring additional samples.<n>AdaQN is theoretically sound and empirically validate it in MuJoCo control problems and Atari $2600 games.
  arXiv  Detail & Related papers  (2024-05-25T11:57:43Z)
• Understanding, Predicting and Better Resolving Q-Value Divergence in Offline-RL [86.0987896274354]
  We first identify a fundamental pattern, self-excitation, as the primary cause of Q-value estimation divergence in offline RL. We then propose a novel Self-Excite Eigenvalue Measure (SEEM) metric to measure the evolving property of Q-network at training. For the first time, our theory can reliably decide whether the training will diverge at an early stage.
  arXiv  Detail & Related papers  (2023-10-06T17:57:44Z)
• AutoRL Hyperparameter Landscapes [69.15927869840918]
  Reinforcement Learning (RL) has shown to be capable of producing impressive results, but its use is limited by the impact of its hyperparameters on performance. We propose an approach to build and analyze these hyperparameter landscapes not just for one point in time but at multiple points in time throughout training. This supports the theory that hyperparameters should be dynamically adjusted during training and shows the potential for more insights on AutoRL problems that can be gained through landscape analyses.
  arXiv  Detail & Related papers  (2023-04-05T12:14:41Z)
• Hyperparameters in Contextual RL are Highly Situational [16.328866317851183]
  Reinforcement Learning (RL) has shown impressive results in games and simulation, but real-world application suffers from its instability under changing environment conditions. We show that the hyper parameters found by HPO methods are not only dependent on the problem at hand, but even on how well the state describes the environment dynamics.
  arXiv  Detail & Related papers  (2022-12-21T09:38:18Z)
• No More Pesky Hyperparameters: Offline Hyperparameter Tuning for RL [28.31529154045046]
  We propose a new approach to tune hyperparameters from offline logs of data. We first learn a model of the environment from the offline data, which we call a calibration model, and then simulate learning in the calibration model. We empirically investigate the method in a variety of settings to identify when it is effective and when it fails.
  arXiv  Detail & Related papers  (2022-05-18T04:26:23Z)
• Sample-Efficient Automated Deep Reinforcement Learning [33.53903358611521]
  We propose a population-based automated RL framework to meta-optimize arbitrary off-policy RL algorithms. By sharing the collected experience across the population, we substantially increase the sample efficiency of the meta-optimization. We demonstrate the capabilities of our sample-efficient AutoRL approach in a case study with the popular TD3 algorithm in the MuJoCo benchmark suite.
  arXiv  Detail & Related papers  (2020-09-03T10:04:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.