NEMOSYS Neural Memory Organization System – Experience-based Neuromorphic Learning for Decision-making and Autonomous Maneuvering at the Edge

Project 1.44 began mid-2025.

Autonomous Ground Vehicles (AGVs) operating in off-road battlefield environments must demonstrate adaptability to dynamic and unpredictable conditions while pursuing mission objectives in novel scenarios. To navigate these complexities, AGVs should integrate prior knowledge of distinct situations alongside an understanding of the temporal and episodic relationships between events and their constituent sub-events. However, the current generation of AGVs, lacking this combination of semantic and episodic intelligence, encounter challenges in adapting to stochastic environments and achieving mission objectives when reliant solely episodic memory mechanisms.

We propose the Neural Memory Organization System (NEMOSYS), a novel interpretation and implementation of semantic and episodic memory systems.. Taking inspiration from the mammalian brain’s semantic and episodic memory systems, NEMOSYS enables AGVs to learn from a structured training curriculum across multiple semantically labelled scenarios. This curriculum will allow them to semantically identify specific battlefield scenarios and respond to temporally related events as distinct episodes. By mimicking the brain’s ability to categorize and recall experiences, NEMOSYS aims to enhance AGV decision-making in dynamic environments. NEMOSYS will enable AGVs to overcome these limitations by integrating semantic memory into the episodic learning framework. This will allow individuals and teams of AGVs to identify specific battlefield scenarios or team formations, enabling them to better understand how to respond to each semantically labeled scenario and its constituent events.

The key research questions (RQs) we are addressing in this project are:

• RQ1: How to develop a brain-inspired algorithm that integrates hippocampus-inspired mechanisms (episodic and semantic memories (ESM)) into the learning framework? Novelty: The development of a neuromorphic algorithm that improves accuracy and minimizes training time and latency by recalling specific scenarios (semantic memory) and the temporal relation betwen their constituent sub-events (episodic memory).
• RQ2: How do ESMs impact learning in SWaP-constrained and dynamic environments? Novelty: The usage of ESM to reduce computational complexity and latency by avoiding redundant training episodes within traditional, non-memory based, learning systems.
• RQ3: How to implement and physically deploy this novel brain-inspired learning and memory system on real-world AGVs? Novelty: Exploring neuromorphic algorithm development for computing hardware available on real-world AGVs.

#1.44