ManimAgent: Self-Evolving Multimodal Agents for Visual Education

Source

Source name: arXiv
Source kind: paper
Authors / organization: Wenjia Jiang, Zongyuan Cai, Yuanhang Shao, Chenru Wang, Boyan Han, Zhixue Song, Keyu Chen, Shengwei An, Xu Yang, Zhou Yang
Venue: arXiv: cs.AI
Published: 2026-06-29
Collected: 2026-06-30
URL: http://arxiv.org/abs/2606.30296v1
PDF: https://arxiv.org/pdf/2606.30296v1
DOI: UNKNOWN

Plain-Language Brief

Researchers present ManimAgent, a self-evolving multimodal agent that uses a dual-channel Episodic Memory Bank to carry reflection experience across tasks in generating Python code using the Manim library for mathematical animations. The memory bank is filled with success rationales and validated failure patterns, which improve performance over time on a fixed-probe evaluation.

What the Document Says

The research focuses on an agent that learns from its failures within a single task but discards the lessons for the next task.
The agent is designed to work on a code-generation task involving scientific paper sections and rendering mathematical animations using the Manim library.
ManimAgent, the proposed agent, stores reflection experience across tasks in a dual-channel Episodic Memory Bank filled from its own task stream, with no human seeds or weight updates.
After each animation converges, a vision-language model scores the rendered keyframes, populating a positive channel M+ and a negative channel M- for success rationales and validated failure patterns respectively.
ManimAgent was tested against various baselines in a fixed-probe evaluation, with human Pass@1 score rising and reflection rounds falling as memory size increases.
The authors will release the code, frozen memory snapshots, and the task stream.

Key Concepts Mentioned

self-evolving agent
multimodal agent
Episodic Memory Bank
Manim library
Python code
mathematical animations
vision-language model
success rationales
validated failure patterns

Claims Made by the Document

ManimAgent improves performance over time.
ManimAgent outperforms various baselines.

Evidence Used

Results from a fixed-probe evaluation against no-memory, matched-budget retrieval-augmented generation, and shuffled-memory baselines.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

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.