- Experience Acquisition: Agents Interact with Environments and Tools to Acquire Experience Trajectories
- Self-Improvement: The Agents Refines its Experiences for Improvement.
- Experience & Policy Integration: Integrating New Experiences into Agents without Catastrophic Forgetting
- Faithfulness: Agent Decisions are Grounded in External Evidence
- Factuality: Maintaining Alignment with World Knowldge, Temporal Events, and Domain-specific Facts
- Consistency Alignment: Ensuring Logical, Factual, and Behavioral Consistency across Agent Trajectories
- Planning & Tool Use: Learning When and How to Explore, Plan, and Invoke Tools
- Evidence & Context Management: Agent-driven Retrieval, Filtering, and Prioritization of Information.
- Iteratvie & Recursive Process: Multi-step, Self-refining Agent Workflows
What We Research
We study how AI agents can continuously evolve through interaction, self-reflection, and experience integration, moving beyond static language models.