MemTransfer: Benchmarking Memory Beyond Matched Experience in Embodied Decision-Making
MemTransfer: Benchmarking Memory Beyond Matched Experience in Embodied Decision-Making
This research introduces MemTransfer, a benchmark assessing six memory representations and a working-memory baseline for embodied agents using a frozen vision-language policy. Evaluated across 100 navigation cases in a simulated warehouse with expert demonstrations, the study compares performance under varying starting poses, route availability, and history relevance. Results indicate that while Full-context and Episodic memory achieve high success at original demonstration starts, they suffer significant drops when conditions change, whereas Summary and Working memory show different robustness profiles depending on the number of demonstrations and task blocking scenarios.
The study evaluates six memory representations against a frozen vision-language policy in a simulated warehouse, demonstrating that high success rates at original demonstration starts do not guarantee robustness to changes in starting pose or route availability.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S3
Reported numbers
loss in success percentage points
48–49
View original evidence
navigation cases across ten task types in a simulated warehouse, with expert demonstrations supplying the history. Three comparisons vary the starting pose, route availability, and amount and task relevance of history. With one demonstration per task, Full-context and Episodic memory reach 95.3% and 100.0% success at the original demonstration start, but lose 48-49 percentage points at a new test start. Summary chang
Open source S3
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
navigation cases across ten task types in a simulated warehouse, with expert demonstrations supplying the history. Three comparisons vary the starting pose, route availability, and amount and task relevance of history. With one demonstration per task, Full-context and Episodic memory reach 95.3% and 100.0% success at the original demonstration start, but lose 48-49 percentage points at a new test start. Summary chang
Open source S3
arXiv:2609.32313v2 Announce Type: replace Abstract: Memory lets an embodied agent reuse past experience, yet retaining useful information does not ensure that the agent can apply it when conditions change. We present MemTransfer, a benchmark comparing six memory representations, a working-memory baseline and five representations of past experience, under a shared frozen vision-language-model policy. It comprises 100
Open source S2
es little between these two test starts, yet with four demonstrations per task it retains a smaller fraction of its unchanged-route success after blocking (39.3%) than Working memory (44.8%) or the two trajectory memories (56-58%). At the new test start, increasing from one to four relevant demonstrations raises Episodic success by 14.3 percentage points, while the other evaluated representations gain no more than 1.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org