MEMOBench

MEMOBench is a process-level memory benchmark for robotic manipulation. It annotates 30 history-dependent tasks with 4,200 executable checkpoints labeling three memory operations: Storage, Update, and Compression. Evaluations show leading VLA policies store information well but fail to update and compress it, reaching only 31.9% average success.

Files in this repository

File	Contents
MEMOBench.zip	The benchmark dataset: 30 demonstration datasets (.hdf5) organized into object/ (5), procedural/ (5), spatial/ (10), and temporal/ (10), with 4,200 executable checkpoints annotated for Storage, Update, and Compression memory operations
assets.zip	Additional object and scene assets required by the LIBERO-based simulator

Usage

hf download SunSeaLucky/MEMOBench MEMOBench.zip --repo-type dataset
hf download SunSeaLucky/MEMOBench assets.zip --repo-type dataset

Extract assets.zip into the LIBERO assets directory. For evaluation code, task configs (.bddl), and detailed instructions, see the GitHub repository.

License

MEMOBench is licensed under CC BY-NC 4.0: free for non-commercial scientific research with attribution; commercial use is not permitted. Third-party code in the accompanying repository (e.g., LIBERO) remains under its own license (MIT).