Search papers, labs, and topics across Lattice.
This paper introduces RingMoClaw, a self-evolving multi-agent framework designed to optimize remote sensing visual interpretation through a closed-loop optimization process. By integrating a research branch, quality-control branch, and a dual-stream dynamic experience bus, the framework autonomously iterates on model design and performance diagnosis, significantly reducing the manual effort typically required. Experimental results show that RingMoClaw outperforms baseline models, achieving a 1.84% improvement in mAP$_{50}$ for object detection and over 40% fewer evolution steps across various tasks, highlighting its effectiveness in continuous model evolution.
RingMoClaw reduces the manual effort in remote sensing model optimization by over 40%, enabling autonomous research iteration that enhances performance across multiple tasks.
Remote sensing visual models have continuously advanced various interpretation tasks. However, the research process behind model improvement still heavily relies on manual expertise, requiring extensive trial-and-error iterations in model design, data processing, and performance diagnosis. Existing agent-based approaches mainly focus on task execution and workflow orchestration, while lacking the capability of autonomous research iteration for continuous performance optimization. To address this issue, we propose RingMoClaw, an experience-inspired self-evolving multi-agent framework for remote sensing visual interpretation. RingMoClaw integrates a research branch, a quality-control branch, and a dual-stream dynamic experience bus to establish a closed-loop optimization process covering strategy generation, experiment execution, independent review, and experience accumulation. The heterogeneous Critic mechanism provides stage-wise diagnosis and feedback, while the dual-stream experience bus incorporates external knowledge and internal experimental experience to guide strategy evolution and eliminate ineffective searches. Extensive experiments on four remote sensing downstream tasks, including object detection, scene classification, semantic segmentation, and change detection, demonstrate the effectiveness and generalization of RingMoClaw. Compared with the corresponding baseline models, RingMoClaw improves performance by 1.84\% mAP$_{50}$ on object detection and achieves consistent gains across the other three tasks, while reducing the required evolution steps by over 40\% compared with existing research automation frameworks. These results suggest that RingMoClaw offers a feasible route from task execution toward continuous research driven model evolution in remote sensing.
