Verified, then logged.
Sun 22 Nov 2026 · 00:21 KST / 15:21 UTC한국어 | ENSearch /

Breaking01 / AI

Spacecraft AI memory scores differ after session restart

AstroMemBench researchers reported no clear advantage for memory systems over retrieval-augmented generation within the same session.

Two monitors in an unoccupied spacecraft simulation lab display different stages of the same fictional satellite maintenance task alongside thumbnails of earlier steps. AI-generated image

AstroMemBench researchers introduced a benchmark on 13 Nov for evaluating spacecraft agents' memory across 2,240 test cases.

The paper covers planetary surface operations, on-orbit servicing and deep-space exploration. It scores actions for task completion, state changes, safety and resource use.

The researchers reported that memory systems exceeded the baseline Pi Agent's average score within the same session. They found no clear advantage over retrieval-augmented generation (RAG), which uses retrieved information to generate answers.

In researcher-published benchmark figures, Mem0 scored 12.25 points higher than RAG after a session restart. The researchers reported that the benefits centred on retaining instructions and reusing previous results more than tracking ongoing state.

Separately, researchers proposed an architecture on 11 Nov that uses request and goal cues to navigate a memory graph, a network of linked information. The design reconstructs the connections between pieces of evidence before generating an answer.

The graph-retrieval paper explicitly states that it reports no benchmark measurements.