The Trinity：面向多模态智能体的能力驱动递归自我改进
原标题：The Trinity:Capability-Driven Recursive Self-Improvement for Multimodal Agents
AI 摘要
论文提出 The Trinity 框架，通过 Creator、Explorer、Evolver 三个角色将环境生成与智能体进化耦合，实现多模态智能体的递归自我改进。Evolver 依据验证过的执行轨迹，把按表面类别组织的任务树重组为按共享能力需求组织的层级能力树，并引入 Capability-Driven Real2Sim2Real 方法。沿能力树进化的 Harness 在 28 个训练基准上平均提升 34 分，在 18 个未见基准上提升 26 分，其中包括两个真实具身任务。
正文节选
The Trinity \setheadertitleThe Trinity: Capability-Driven Recursive Self-Improvement for Multimodal Agents Capability-Driven Recursive Self-Improvement for Multimodal Agents Abstract Multimodal intelligence can be viewed as human-guided co-evolution between worlds and agents through interaction. However, scaling this process faces two bottlenecks: constructing heterogeneous, verifiable environments is costly, and manual Harness iteration cannot keep pace with growing task diversity and complexit