Axon

Axon's development of the Fleet 3 in-car camera system — a mission-critical device deployed in active law enforcement vehicles and subject to continuous vibration, temperature swings, and mechanical shock — coincided with a high-risk combination of new overseas contract manufacturers in Taiwan and China and fresh factory bring-ups. COVID-19 travel restrictions eliminated on-site engineering support starting January 2020, leaving teams distributed across the US, Finland, and Vietnam with no physical presence at manufacturing sites. Functional tests couldn't catch every defect class; components like sealing gaskets and fasteners required visual inspection that wasn't scalable across all units. Any defect escape in a safety-critical product carried serious downstream consequences across the judicial system and law enforcement community.

Axon deployed Instrumental's AI-powered computer vision platform across its development and production lines at contract manufacturing sites. Unlike traditional automated optical inspection (AOI) systems that require manual programming for each defect type, Instrumental's anomaly detection system learned from image data without pre-defining defect signatures — effectively operating as a remote product design engineer reviewing each unit. From day one of each build phase, the system captured a full image record of every unit, enabling the globally distributed engineering team to perform remote failure analysis asynchronously. This eliminated the dependency on travel for quality oversight and gave NPI engineers in the US and Finland direct visual access to units being built in Taiwan and China, with complete traceability across builds.

Instrumental's platform delivered measurable impact across every phase of Fleet 3 development. The headline outcome was the discovery of more than 20 distinct unknown defect types — anomalies that standardized functional tests would not have surfaced and that carried escape risk into finished product. Defect rates fell 6% from EVT1 to DVT, compressing the quality improvement curve across development phases and avoiding rework on dozens of units that would otherwise have required manual re-inspection. In EVT2 alone, the team recovered 35 engineering hours by focusing effort on highest-priority issues rather than manually triaging every anomaly. Traceability across builds also improved substantially, enabling rapid cross-build comparisons backed by historical image data.

• 20+ unknown issue types identified before escape
• 6% defect rate reduction from EVT1 to DVT
• 35 engineering hours saved in EVT2 alone
• Full unit-level traceability established across build phases

• AI-based anomaly detection surfaces defect classes that functional test coverage structurally cannot — making it a complement, not a replacement, for existing test regimes in electronics NPI.
• Remote visual inspection infrastructure is now a risk mitigation requirement for any program relying on overseas contract manufacturers, particularly when travel cannot be assumed.
• Deploying during EVT rather than waiting for mass production maximizes ROI — defects found early prevent exponential rework costs in later builds.
• Platforms that require no defect pre-programming reduce deployment friction significantly; teams can generate value from day one without a lengthy configuration phase.
• Cross-build image traceability accelerates root cause analysis and prevents recurring defects from resurfacing in subsequent build phases.