EvolvingBrain: An Adaptive Skill System for Multi-modal Agents Self-Evolution in Open Worlds
Quick Answer
EvolvingBrain introduces an adaptive skill system for multi-modal agents, enhancing self-evolution in open worlds.
Quick Take
It features Stratified Skill Self-Evolution and State-Aware Adaptive Execution, improving task success rates in VAB-Minecraft and DiscoveryWorld. The system promotes exploration and knowledge utilization based on current states, demonstrating significant generalization capabilities.
Key Points
- EvolvingBrain enhances multi-modal agents' self-evolution capabilities.
- It employs a Stratified Skill Architecture for knowledge organization.
- Experiments show improved success rates in VAB-Minecraft and DiscoveryWorld.
- The system utilizes structured memory for state-aware skill execution.
- Code will be publicly released in the camera-ready version.
DeepSignal Analysis
What happened
EvolvingBrain is a proposed adaptive skill system aimed at enhancing the self-evolution of multi-modal agents in open-world environments. It includes two main components: Stratified Skill Self-Evolution and State-Aware Adaptive Execution, which together improve task success rates in specific benchmarks like VAB-Minecraft and DiscoveryWorld.
Key evidence
- EvolvingBrain introduces Stratified Skill Self-Evolution, which extracts patterns from model-specific trajectories to evolve agent skills through interactions with the open world.
- The system's State-Aware Adaptive Execution uses long-horizon interactions to guide agents in utilizing knowledge based on their current state.
- Experiments conducted on VAB-Minecraft and DiscoveryWorld show that EvolvingBrain significantly improves success rates on open-world tasks.
Why it matters
The development of EvolvingBrain addresses the limitations of existing multi-modal agents, which struggle with self-evolution and effective knowledge utilization. By enhancing agents' ability to adapt and learn from their environments, this system could lead to more capable AI in complex scenarios, potentially impacting various applications in AI-driven exploration and problem-solving.
Paper Resources
📖 Reader Mode
~2 min readAuthors:Runxuan Liu, Jingqi Shi, Baoxin Wang, Honghong Zhao, Haoyuan Xu, Yueqing Sun, Jiafeng Liang, Tao He, Dayong Wu, Qi Gu, Shijin Wang, Ming Liu, Bing Qin
Abstract:Multi-modal agents have made significant progress in interacting with open worlds. However, their ability to self-evolve remains limited. Supervised fine-tuning enables imitation but not active experience organization, while outcome-based reinforcement learning promotes exploration yet underuses intermediate trajectory information. Therefore, we propose EvolvingBrain, which uses an adaptive skill system as the brain for agent self-evolution, consisting of two following components. Stratified Skill Self-Evolution: extracts Procedural Patterns and Problematic Patterns from model-specific trajectories, summarizes experience by generating skills, and evolves through multiple interactions with the open world, ultimately forming a Stratified Skill Architecture consisting of Domain, Plan, and Action/Trap Skills, which provides the agent with a multi-granularity knowledge repository and a novel evolution approach. State-Aware Adaptive Execution: summarizes long-horizon interactions into Structured Abstractive Memory and uses it to guide the State-Aware Skill Fusion Strategy, enabling the agent to utilize knowledge according to its current state. Experiments on VAB-Minecraft and DiscoveryWorld demonstrate that EvolvingBrain improves success rates on open-world tasks and exhibits generalization capabilities. Our code is provided in the supplementary materials and will be publicly released in the camera-ready version.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.20786 [cs.AI] |
| (or arXiv:2610.20786v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.20786 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Runxuan Liu [view email]
[v1]
Wed, 14 Oct 2026 18:30:07 UTC (5,223 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →VEPA: Verified Evolving Formal Policy for Lifelong Agent Guardrails
VEPA introduces a framework for verified evolving formal policies for LLM agents, enhancing safety by adapting guardrails based on deployment feedback. It outperforms existing methods across benchmarks like ASSEBench-Safety and ASSEBench-Security, ensuring deterministic policy enforcement even under challenging conditions.