Multi modal agents adapt skills for open world tasks
EvolvingBrain: An Adaptive Skill System for Multi-modal Agents Self-Evolution in Open Worlds
Artificial Intelligence
Summary
Many AI agents can now interact with complex virtual worlds, but they struggle to learn and improve on their own over time. This paper presents EvolvingBrain, a new way for agents to create and refine skills based on their own experiences and challenges. The system organizes knowledge at different levels and remembers situations to decide which skills to use next. Tests in game-like environments showed that agents with EvolvingBrain performed better and adapted more easily to new tasks. The authors made their code available for others to build on this approach.
What this means in practice
- •For game developers: Create AI characters that adaptively learn new skills and strategies by interacting with complex game worlds without additional programming.$Commercial implications: Enables development of more adaptive and autonomous non-player characters that enhance player engagement and reduce manual AI design.
- •For roboticists: Build robots capable of evolving task-specific skills autonomously by summarizing experiences and adapting actions to changing environments.
Authors
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.