PhGPO: Pheromone-Guided Policy Optimization for Long-Horizon Tool Planning

arXiv:2602.13691 · cs.AI · Submitted 2026-02-14 · Read on arXiv

cs.AI

Submitted: 2026-02-14

Updated: 2026-10-01

Comments: NeurIPS 2026 Poster

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Large Language Model (LLM) — 1× in this paper · explained in 3 episodes

Sources

• AgentOrchestra: Orchestrating Multi-Agent Intelligence with the Tool-Environment-Agent(TEA) Protocol
• MCP-Universe: Benchmarking Large Language Models with Real-World Model Context Protocol Servers
• Learning from All-Failed Groups via First-Visit Coverage in Long-Horizon Agentic Tasks
• Learning Robust Reasoning through Guided Adversarial Self-Play
• Phase-Aware Mixture of Experts for Agentic Reinforcement Learning
• Alita: Generalist Agent Enabling Scalable Agentic Reasoning with Minimal Predefinition and Maximal Self-Evolution
• From Code Foundation Models to Agents and Applications: A Comprehensive Survey and Practical Guide to Code Intelligence
• The Tool Decathlon: Benchmarking Language Agents for Diverse, Realistic, and Long-Horizon Task Execution
• TOUCAN: Synthesizing 1.5M Tool-Agentic Data from Real-World MCP Environments
• TRAJECT-Bench:A Trajectory-Aware Benchmark for Evaluating Agentic Tool Use
• ToRL: Scaling Tool-Integrated RL
• ARPO:End-to-End Policy Optimization for GUI Agents with Experience Replay
• SALT: Step-level Advantage Assignment for Long-horizon Agents via Trajectory Graph
• Pheromone-based Learning of Optimal Reasoning Paths
• Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models
• ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs
• StableToolBench: Towards Stable Large-Scale Benchmarking on Tool Learning of Large Language Models
• API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs
• Online-Optimized RAG for Tool Use and Function Calling
• ToolNet: Connecting Large Language Models with Massive Tools via Tool Graph

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• Agents' Last Exam
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems