In mission-critical scenarios such as disaster inspection and search-and-rescue, communication-limited robots must make reliable onboard decisions without access to remote operators or high-capacity reasoning services. Episodic memory reuse serves as an attractive low-cost fallback, but retrieval similarity does not guarantee execution validity; a retrieved action might align with the current context but could be unsafe due to topology changes, insufficient battery margins, or unreliable prior outcomes. We term these high-similarity but execution-invalid episodes as memory traps. This creates a safety-efficiency design space where similarity-only reuse minimizes fallback costs but may be unsafe, while always invoking local reasoning enhances safety at high computational and energy costs. This paper introduces MemoGuard, a lightweight adaptive runtime that validates episodic memories against topology, resource, and outcome contracts before reuse, invoking fallback only when validation fails. In a graph-based corridor-inspection simulator, MemoGuard reduces battery safety violations by 76.6% compared to similarity-only top-1 reuse while also reducing fallback calls by 21.4% compared to always reasoning. On an NVIDIA Jetson AGX Xavier with local llama3.2:3b fallback reasoning, this corresponds to 3.67 seconds and 36.97 Joules of avoided fallback reasoning overhead per trial. We open-source MemoGuard at https://github.com/hetheiin/memoguard.
Blogger's Review: MemoGuard presents an innovative solution to the decision-making challenges faced by robots in communication-limited environments by effectively validating the safety of episodic memories, significantly enhancing navigation safety and efficiency. Its open-source nature also provides a solid foundation for further research, making it a noteworthy development in the field.