ManimAgent: Self-Evolving Multimodal Agents for Visual Education
This work addresses the problem of wasted reflection experience in LLM-based agents for code generation, offering a memory mechanism that improves efficiency over no-memory and RAG baselines.
ManimAgent introduces a self-evolving multimodal agent that retains cross-task reflection experience via a dual-channel Episodic Memory Bank, achieving improved Pass@1 and reduced reflection rounds on a Manim code-generation task without weight updates or human seeds.
Multi-round reflection lets agents built on large language models recover from failures within a single task, but each task remains an isolated episode: lessons learned across many reflection rounds on one task are discarded before the next begins. We study this gap on a code-generation task: from a scientific paper section, the agent writes Python in the open-source Manim library to render a mathematical animation. We present ManimAgent, a self-evolving multimodal agent that carries reflection experience across tasks through a dual-channel Episodic Memory Bank grown entirely from its own task stream, with no weight updates and no human seeds. After each animation converges, a vision-language model scores the rendered keyframes; the resulting signals populate a positive channel M+ that stores success rationales as soft Reference Examples, and a negative channel M- that stores validated failure patterns as hard Known Pitfalls. On a fixed-probe evaluation against no-memory, matched-budget retrieval-augmented generation, and shuffled-memory baselines, blind human Pass@1 rises and reflection rounds fall as memory size grows. We will release the code, frozen memory snapshots, and the task stream.