MemGuide is the training dataset for MemAdapter,
a plug-and-play module that converts retrieved memories into structured planning guidance for
Vision-Language Model (VLM) embodied agents.
Each record pairs a task instruction and retrieved memories (spatial, episodic, and semantic)
with structured planning guidance produced by a frontier LLM and filtered by behavioral consensus.
The dataset is used to fine-tune the MemAdapter (Qwen3-14B +… See the full description on the dataset page:
https://huggingface.co/datasets/NMThuan032k/MemGuide.
