S3Mem: Structured Spatiotemporal Scene-Event Memory for Long-Horizon Interactive Question Answering

In three linesS3MEM introduces a structured scene-event episodic memory framework for long-horizon interactive agents. The system structures trajectories into organized memory units and uses anchor-sensitive retrieval to improve spatiotemporal question answering. Evaluated on Crafter, Jericho, SciWorld, and ALFWorld, S3MEM outperforms Vanilla RAG and Graph-NoReader in accuracy while using fewer evidence tokens.

Summary generated by Claude — human-verified