pith. sign in

arxiv: 2510.08558 · v3 · pith:ATSZCHBVnew · submitted 2025-10-09 · 💻 cs.AI · cs.CL· cs.IR· cs.LG

Agent Learning via Early Experience

classification 💻 cs.AI cs.CLcs.IRcs.LG
keywords experienceagentagentsdataearlylearningenvironmentsimprove
0
0 comments X
read the original abstract

A long-term goal of language agents is to learn and improve through their own experience, ultimately outperforming humans in complex, real-world tasks. However, training agents from experience data with reinforcement learning remains difficult in many environments, which either lack verifiable rewards (e.g., websites) or require inefficient long-horizon rollouts (e.g., multi-turn tool use). As a result, most current agents rely on supervised fine-tuning on expert data, which is challenging to scale and generalizes poorly. This limitation stems from the nature of expert demonstrations: they capture only a narrow range of scenarios, and expose the agent to limited environment diversity. We address this limitation with a middle-ground paradigm we call early experience: interaction data generated by the agent's own actions, where the resulting future states serve as supervision without reward signals. Within this paradigm, we study two strategies of using such data: (1) implicit world modeling, which uses collected states to ground the policy in environment dynamics; and (2) self-reflection, where the agent learns from its suboptimal actions to improve reasoning and decision-making. Evaluation across eight diverse environments and multiple model families shows that our approaches consistently improve effectiveness and out-of-domain generalization, highlighting the value of early experience. Moreover, in environments with verifiable rewards, our results provide promising signals that early experience offers a strong foundation for subsequent reinforcement learning, making it a practical bridge between imitation learning and fully experience-driven agents.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 29 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Beyond Next-Observation Prediction: Agent-Authored World Modeling for Sequential Decision Making

    cs.CL 2026-06 unverdicted novelty 7.0

    AAWM builds training targets for world models by retrieving and synthesizing transition evidence based on the policy's self-identified decision needs at each state.

  2. Co-Evolving Skill Generation and Policy Optimization

    cs.CL 2026-06 unverdicted novelty 7.0

    Framework estimates context-dependent marginal utility of candidate skills via reward gaps in matched base vs. skill-augmented rollouts to filter skills and co-train policy as generator.

  3. EXG: Self-Evolving Agents with Experience Graphs

    cs.AI 2026-05 unverdicted novelty 7.0

    EXG is an experience graph framework for self-evolving LLM agents that supports online real-time growth and offline reuse to enhance solution quality and efficiency on code generation and reasoning benchmarks.

  4. Revisiting the Travel Planning Capabilities of Large Language Models

    cs.AI 2026-05 unverdicted novelty 7.0

    LLMs extract explicit constraints effectively but struggle with implicit open-world requirements, structural biases in plans, and ineffective self-correction during travel planning.

  5. Reference-Sampled Boltzmann Projection for KL-Regularized RLVR: Target-Matched Weighted SFT, Finite One-Shot Gaps, and Policy Mirror Descent

    cs.LG 2026-05 unverdicted novelty 7.0

    Reference-sampled weighted SFT with prompt-normalized Boltzmann weights induces the same policy as fixed-reference KL-regularized RLVR, with BOLT as the estimator and a finite one-shot error decomposition separating c...

  6. Where Do CoT Training Gains Land in LLM based Agents?

    cs.AI 2026-06 unverdicted novelty 6.0

    CoT training in LLM agents improves prompt-action quality more than the advantage of generated reasoning, and selectively masking action supervision improves out-of-domain generalization.

  7. The Interplay of Harness Design and Post-Training in LLM Agents

    cs.LG 2026-06 unverdicted novelty 6.0

    Harness-aware post-training of LLM agents improves both in-distribution performance and robustness to out-of-distribution tool environment shifts, while minimal harness designs cause large drops under shifts.

  8. Qwen-AgentWorld: Language World Models for General Agents

    cs.CL 2026-06 unverdicted novelty 6.0

    Qwen-AgentWorld are language world models that simulate multi-domain agent environments and boost general agent capabilities via decoupled RL simulation and unified foundation model training.

  9. Escaping the Self-Confirmation Trap: An Execute-Distill-Verify Paradigm for Agentic Experience Learning

    cs.CL 2026-06 unverdicted novelty 6.0

    EDV decouples execution, distillation by a third-party agent, and consensus verification to filter erroneous trajectories in LLM agent experience learning, outperforming baselines on tau2-bench, Mind2Web, and MMTB.

  10. Demo2Tutorial: From Human Experience to Multimodal Software Tutorials

    cs.CV 2026-06 unverdicted novelty 6.0

    Demo2Tutorial distills human screen recordings into hierarchical image-text tutorials that outperform human-authored ones on a documentation-derived benchmark and improve downstream human task speed and GUI-agent planning.

  11. Policy and World Modeling Co-Training for Language Agents

    cs.LG 2026-06 unverdicted novelty 6.0

    PaW co-trains policy and world modeling on standard RL rollouts using action-entropy data selection, noise-tolerant loss, and reward-adaptive balancing, yielding consistent gains on three agent benchmarks.

  12. COMAP: Co-Evolving World Models and Agent Policies for LLM Agents

    cs.AI 2026-06 unverdicted novelty 6.0

    COMAP co-evolves textual world models and agent policies for LLMs through on-policy self-distillation, yielding up to 16.75% relative gains on embodied planning, web navigation, and tool-use tasks.

  13. Enhancing LLM Metacognition via Cognitive Pairwise Training

    cs.LG 2026-05 unverdicted novelty 6.0

    CPT is introduced as a pairwise reasoning-trace comparison stage that improves the reasoning-metacognition trade-off over standard SFT+RL pipelines across model scales.

  14. ECHO: Terminal Agents Learn World Models for Free

    cs.LG 2026-05 unverdicted novelty 6.0

    ECHO is a hybrid RL objective that trains agents to predict environment observation tokens from their actions, doubling GRPO pass@1 on TerminalBench-2.0 while improving dynamics prediction on held-out trajectories.

  15. Differentiable Mixture-of-Agents Incentivizes Swarm Intelligence of Large Language Models

    cs.LG 2026-05 unverdicted novelty 6.0

    DMoA is a differentiable multi-agent framework for LLMs that uses recurrent context-aware routing and predictive entropy for test-time adaptation, claiming SOTA results on 9 benchmarks with efficiency and robustness.

  16. MAP: A Map-then-Act Paradigm for Long-Horizon Interactive Agent Reasoning

    cs.AI 2026-05 unverdicted novelty 6.0

    MAP improves LLM agent reasoning by constructing a structured cognitive map of the environment before task execution, yielding performance gains on benchmarks like ARC-AGI-3 and superior training data via the new MAP-...

  17. Learning Agent Routing From Early Experience

    cs.CL 2026-05 unverdicted novelty 6.0

    BoundaryRouter routes queries to LLM or agent using early experience memory from a seed set, cutting inference time 60.6% versus always using agents and raising performance 28.6% versus always using direct LLM inference.

  18. From History to State: Constant-Context Skill Learning for LLM Agents

    cs.AI 2026-05 unverdicted novelty 6.0

    Constant-context skill learning trains reusable task-family modules for LLM agents using a deterministic state block for progress tracking and subgoal rewards, achieving 89.6% unseen success on ALFWorld, 76.8% on WebS...

  19. CASCADE: Case-Based Continual Adaptation for Large Language Models During Deployment

    cs.AI 2026-05 unverdicted novelty 6.0

    CASCADE enables LLMs to continually adapt at deployment via case-based episodic memory and contextual bandits, improving macro-averaged success by 20.9% over zero-shot on 16 tasks spanning medicine, law, code, and robotics.

  20. Training LLM Agents for Spontaneous, Reward-Free Self-Evolution via World Knowledge Exploration

    cs.AI 2026-04 unverdicted novelty 6.0

    LLM agents trained with a task-success reward on self-generated knowledge can spontaneously explore and adapt to new environments without any rewards or instructions at inference, yielding 20% gains on web tasks and a...

  21. HEALing Entropy Collapse: Enhancing Exploration in Few-Shot RLVR via Hybrid-Domain Entropy Dynamics Alignment

    cs.LG 2026-04 unverdicted novelty 6.0

    HEAL mitigates entropy collapse in few-shot RLVR by selectively adding general-domain data and aligning trajectory-level entropy dynamics, matching full-shot performance with 32 target samples.

  22. Time is Not a Label: Continuous Phase Rotation for Temporal Knowledge Graphs and Agentic Memory

    cs.CL 2026-04 unverdicted novelty 6.0

    RoMem uses a Semantic Speed Gate to assign volatility to relations and continuous phase rotation to shadow obsolete facts in complex space, delivering SOTA temporal KG completion and 2-3x gains on agentic memory benchmarks.

  23. Agentic Learner with Grow-and-Refine Multimodal Semantic Memory

    cs.AI 2025-11 unverdicted novelty 6.0

    ViLoMem is a dual-stream grow-and-refine memory system that separates visual and logical error patterns in MLLMs to improve pass@1 accuracy and reduce repeated mistakes across six multimodal benchmarks.

  24. Self-Evolving World Models for LLM Agent Planning

    cs.AI 2026-06 unverdicted novelty 5.0

    WorldEvolver uses episodic memory, semantic memory, and selective foresight to self-evolve world models at test time, achieving top prediction accuracy and agent success on ALFWorld and ScienceWorld benchmarks.

  25. Internalizing the Future: A Unified Agentic Training Paradigm for World Model Planning

    cs.AI 2026-06 unverdicted novelty 5.0

    A three-stage training pipeline internalizes world-model simulation and success estimation in LLM agents for improved planning on search and math tasks.

  26. Differentiable Mixture-of-Agents Incentivizes Swarm Intelligence of Large Language Models

    cs.LG 2026-05 unverdicted novelty 5.0

    DMoA is a differentiable multi-agent LLM framework with recurrent context-aware routing and predictive entropy self-supervision that claims SOTA results on 9 benchmarks through elastic agent collaboration.

  27. Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering

    cs.SE 2026-04 accept novelty 5.0

    LLM agent progress depends on externalizing cognitive functions into memory, skills, protocols, and harness engineering that coordinates them reliably.

  28. UI-Oceanus: Scaling GUI Agents with Synthetic Environmental Dynamics

    cs.LG 2026-02 unverdicted novelty 5.0

    UI-Oceanus shows that continual pre-training on forward dynamics predictions from synthetic GUI exploration improves agent success rates by 7% offline and 16.8% online, with gains scaling by data volume.

  29. Meta-Tool: Efficient Few-Shot Tool Adaptation for Small Language Models

    cs.CL 2026-04 unverdicted novelty 4.0

    A 3B model with few-shot prompting reaches 79.7% of GPT-5 tool-use performance while a hypernetwork adaptation adds zero measurable benefit across four benchmarks.