Aligning Agents via Planning: A Benchmark for Trajectory-Level Reward Modeling

• URL: http://arxiv.org/abs/2604.08178v1
• Date: Thu, 09 Apr 2026 12:35:06 GMT
• Title: Aligning Agents via Planning: A Benchmark for Trajectory-Level Reward Modeling
• Abstract summary: Plan-RewardBench is a trajectory-level preference benchmark designed to evaluate how well judges distinguish preferred versus distractor agent trajectories.<n>Plan-RewardBench covers four representative task families -- (i) Safety Refusal, (ii) Tool-Irrelevance / Unavailability, (iii) Complex Planning, and (iv) Robust Error Recovery.
• Score: 19.766968596602457
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: In classical Reinforcement Learning from Human Feedback (RLHF), Reward Models (RMs) serve as the fundamental signal provider for model alignment. As Large Language Models evolve into agentic systems capable of autonomous tool invocation and complex reasoning, the paradigm of reward modeling faces unprecedented challenges--most notably, the lack of benchmarks specifically designed to assess RM capabilities within tool-integrated environments. To address this gap, we present Plan-RewardBench, a trajectory-level preference benchmark designed to evaluate how well judges distinguish preferred versus distractor agent trajectories in complex tool-using scenarios. Plan-RewardBench covers four representative task families -- (i) Safety Refusal, (ii) Tool-Irrelevance / Unavailability, (iii) Complex Planning, and (iv) Robust Error Recovery -- comprising validated positive trajectories and confusable hard negatives constructed via multi-model natural rollouts, rule-based perturbations, and minimal-edit LLM perturbations. We benchmark representative RMs (generative, discriminative, and LLM-as-Judge) under a unified pairwise protocol, reporting accuracy trends across varying trajectory lengths and task categories. Furthermore, we provide diagnostic analyses of prevalent failure modes. Our results reveal that all three evaluator families face substantial challenges, with performance degrading sharply on long-horizon trajectories, underscoring the necessity for specialized training in agentic, trajectory-level reward modeling. Ultimately, Plan-RewardBench aims to serve as both a practical evaluation suite and a reusable blueprint for constructing agentic planning preference data.

Related papers

• Internalizing the Future: A Unified Agentic Training Paradigm for World Model Planning [32.26884129316165]
  Large language model (LLM) agents have demonstrated strong capability in sequential decision-making.<n>Unlike humans who employ "what-if" reasoning to evaluate potential plans before commitment, standard agents lack an internal world model to simulate future outcomes.<n>We propose to internalize future-aware planning by training a single autoregressive model to verbalize both a prospective state rollout and a plan-conditioned success estimate.
  arXiv  Detail & Related papers  (2026-06-25T19:05:44Z)
• On Predicting the Post-training Potential of Pre-trained LLMs [60.07459271263409]
  We introduce a new task of predicting post-training potential - forecasting a base model's performance before post-training.<n>We propose RuDE, a unified framework that bypasses the generation gap of base models by leveraging response discrimination.<n>Experiments demonstrate a correlation greater than 90% with post-training performance.
  arXiv  Detail & Related papers  (2026-05-12T11:33:49Z)
• TopoCurate:Modeling Interaction Topology for Tool-Use Agent Training [53.93696896939915]
  Training tool-use agents typically rely on Supervised Fine-Tuning (SFT) on successful trajectories and Reinforcement Learning (RL) on pass-rate-selected tasks.<n>We propose TopoCurate, an interaction-aware framework that projects multi-trial rollouts from the same task into a unified semantic quotient topology.<n>TopoCurate achieves consistent gains of 4.2% (SFT) and 6.9% (RL) over state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-03-02T10:38:54Z)
• MagicAgent: Towards Generalized Agent Planning [73.21129030631421]
  We present textbfMagicAgent, a series of foundation models specifically designed for generalized agent planning.<n>We introduce a lightweight and scalable synthetic data framework that generates high-quality trajectories across diverse planning tasks.<n>We show that MagicAgent-32B and MagicAgent-30B-A3B achieve superior performance across diverse open-source benchmarks.
  arXiv  Detail & Related papers  (2026-02-22T01:39:16Z)
• PILOT: Planning via Internalized Latent Optimization Trajectories for Large Language Models [51.43746425777865]
  Large Language Models (LLMs) often lack the capacity to formulate global strategies, leading to error propagation in long-horizon tasks.<n>We propose PILOT, a framework designed to internalize the strategic oversight of large models into intrinsic Latent Guidance.
  arXiv  Detail & Related papers  (2026-01-07T12:38:56Z)
• SCRIBE: Structured Mid-Level Supervision for Tool-Using Language Models [10.04930078540686]
  SCRIBE is a reinforcement learning framework that intervenes at a novel mid-level abstraction.<n>It achieves state-of-the-art performance across a range of reasoning and tool-use benchmarks.
  arXiv  Detail & Related papers  (2026-01-07T03:49:48Z)
• Spatiotemporal Forecasting as Planning: A Model-Based Reinforcement Learning Approach with Generative World Models [45.523937630646394]
  We propose SFP Forecasting as Planning (SFP), a new paradigm in Model Based Reinforcement Learning.<n>SFP constructs a novel World Model to simulate diverse high-temporal future states, enabling an "imagination-based" environmental simulation.
  arXiv  Detail & Related papers  (2025-10-05T03:57:38Z)
• PilotRL: Training Language Model Agents via Global Planning-Guided Progressive Reinforcement Learning [19.480628850056522]
  Large Language Models (LLMs) have shown remarkable advancements in tackling agent-oriented tasks.<n>Current approaches predominantly rely on supervised fine-tuning, which often leads models to memorize established task completion trajectories.<n>We introduce an adaptive global plan-based agent paradigm AdaPlan, aiming to synergize high-level explicit guidance with execution.
  arXiv  Detail & Related papers  (2025-08-01T06:17:11Z)
• Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models [79.2162092822111]
  We systematically evaluate reinforcement learning (RL) and control-based methods on a suite of navigation tasks.<n>We employ a latent dynamics model using the Joint Embedding Predictive Architecture (JEPA) and employ it for planning.<n>Our results show that model-free RL benefits most from large amounts of high-quality data, whereas model-based planning generalizes better to unseen layouts.
  arXiv  Detail & Related papers  (2025-02-20T18:39:41Z)
• InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling [66.3072381478251]
  Reward hacking, also termed reward overoptimization, remains a critical challenge. We propose a framework for reward modeling, namely InfoRM, by introducing a variational information bottleneck objective. We show that InfoRM's overoptimization detection mechanism is not only effective but also robust across a broad range of datasets.
  arXiv  Detail & Related papers  (2024-02-14T17:49:07Z)
• When Demonstrations Meet Generative World Models: A Maximum Likelihood Framework for Offline Inverse Reinforcement Learning [62.00672284480755]
  This paper aims to recover the structure of rewards and environment dynamics that underlie observed actions in a fixed, finite set of demonstrations from an expert agent. Accurate models of expertise in executing a task has applications in safety-sensitive applications such as clinical decision making and autonomous driving.
  arXiv  Detail & Related papers  (2023-02-15T04:14:20Z)

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.