Anomaly Detection with a Robotic Edge Device

Watch the video

Military / Coalition Issue

A tactical edge is a dynamically changing environment. Any anomalous event can potentially have massive financial and operational repercussions for military. In addition, human life can be at risk. This work focuses on anomaly detection using a robotic edge device which senses multimodal information from the environment over time to detect changes in environment and flag anomalies.

Core idea and key achievements

Developed an initial multi-modal anomaly detection methodology which aligns information from multiple modalities (e.g., visual, thermal, audio etc.) and uses a temporal modeling methodology to detect anomalous events. The methodology is unsupervised and uses changes in patterns over time for anomaly detection.

Implications for Defence

Anomaly detection at the tactical edge is a particularly important problem for defence purposes. A roaming edge-device such as a robot can be immensely useful for such an application as it can detect anomalies in places where it is risky to send people and can spot things obscured from aerial view (e.g. under shrubbery, exploded buildings etc.). Subtle anomalies which would otherwise elude the human eye can also be detected using advanced sensing (e.g. IR imaging). Anomalies can appear in many forms such as (1) presence of suspicious devices, (2) soldiers who need assistance (during a conflict), or (3) overheating military equipment. Hence, analysis of multimodal data is essential.

An example pattern for the military would be: during normal period, a robotic edge-device roams around a tactical edge facility and captures multimodal information (visual, audio, thermal etc.) at specific places. This is sent to an edge node which incrementally updates an AI model of what is “normal” at these locations. If at a future time point, some anomalous event occurs, the multimodal anomaly detection module flags it and raises an exception in the defence enterprise asset management system so that next steps are taken.

image info

Readiness & alternative Defence uses

The capability for version 1 of an edge anomaly detection system that incorporates the described functionality is currently in progress within IBM Research. It incorporates and extends DAIS research in model selection, model management, and model fingerprinting. It also incorporates some non-DAIS IBM research and integrates with existing edge deployment and management software. Further progress has been set out to provide a vision for the next stage of anomaly detection system in the near future and a road map towards version 2. Various parts of the capabilities of the described system are at different maturity levels that approximate to TRL levels 3, 4 and 5. Thus, ready to prototype a military Edge AI production facility, and experiment with a range of potential usage patterns.

Resources and references