Spark: Strategic Policy-Aware Exploration via Dynamic Branching for Long-Horizon Agentic Learning
This addresses the problem of inefficient resource allocation in reinforcement learning for long-horizon agentic learning, offering a novel method to improve sample efficiency and generalization, though it appears incremental as it builds on existing exploration strategies.
The paper tackles the challenge of training large language models as agents for long-horizon tasks by proposing Spark, a framework that selectively branches at critical decision states for resource-efficient exploration, achieving superior success rates with significantly fewer training samples and robust generalization in experiments.
Reinforcement learning has empowered large language models to act as intelligent agents, yet training them for long-horizon tasks remains challenging due to the scarcity of high-quality trajectories, especially under limited resources. Existing methods typically scale up rollout sizes and indiscriminately allocate computational resources among intermediate steps. Such attempts inherently waste substantial computation budget on trivial steps while failing to guarantee sample quality. To address this, we propose \textbf{Spark} (\textbf{S}trategic \textbf{P}olicy-\textbf{A}ware explo\textbf{R}ation via \textbf{K}ey-state dynamic branching), a novel framework that selectively branches at critical decision states for resource-efficient exploration. Our key insight is to activate adaptive branching exploration at critical decision points to probe promising trajectories, thereby achieving precise resource allocation that prioritizes sampling quality over blind coverage. This design leverages the agent's intrinsic decision-making signals to reduce dependence on human priors, enabling the agent to autonomously expand exploration and achieve stronger generalization. Experiments across diverse tasks (e.g., embodied planning), demonstrate that \textsc{Spark} achieves superior success rates with significantly fewer training samples, exhibiting robust generalization even in unseen scenarios.