Overview
The RedTeam Defence-Tech Hackathon put ten teams at the Lassonde School of Engineering, York University, against a single hard problem: navigate when the satellite fix is gone and the link back to the operator is gone with it. GPS denial is not a hypothetical in contested airspace, it is the normal operating condition, and a vehicle that stops being useful the moment it loses signal is not an autonomous vehicle.
The Alpha Nova took 1st place ($10,000) and Best Presenter ($1,000), flying a drone that navigates on computer vision alone. The work is run through AN Labs, our Canadian R&D division.
The Challenge
- Hold an accurate position estimate with no GNSS fix available
- Keep operating through loss of the communications link to the ground station, rather than failing safe and stopping
- Run the whole decision loop on constrained onboard compute, in real time
- Survive adversarial conditions rather than a clean test environment
- Compete against ten teams, and defend the engineering under judging
Our Approach
If the satellite fix is gone, the aircraft has to work out where it is from what it can see. We built the navigation around onboard computer vision, which is the same discipline behind our open-pit blast analysis for Côté Gold and our wearable-free fall detection in VanGuardian, applied here to a drone that has to stay useful with no GPS and no operator.
- Vision-Based Positioning: position estimated from the downward and forward camera feeds, tracking visual features across frames to recover motion rather than reading it off a satellite
- Visual-Inertial Fusion: camera data fused with the inertial unit, so short visual dropouts over featureless ground are bridged by inertial data and drift stays bounded instead of compounding
- Perception Under Adversarial Conditions: a feed that is degraded, repetitive or deliberately interfered with is the expected input, not the failure case
- Inference on the Airframe: the vision pipeline runs on the drone's own compute. In a comms-denied scenario there is no ground station to offload to, so anything requiring a round trip is already useless
- Predictable Degradation: confidence in the position estimate is tracked explicitly, so the aircraft knows when its own fix is weakening and behaves as a defined state rather than an error path
- The Reasoning, Not Just the Demo: presenting why the design holds up, which is what took Best Presenter alongside the win
Impact
10
Teams Competed Against
- 1st place in the GPS and communications-denied navigation challenge
- Best Presenter, awarded for the engineering case behind the build
- Validated our edge autonomy work against an adversarial, judged problem rather than a benchmark
- Reinforced AN Labs as a Canadian, single-jurisdiction partner for dual-use autonomy work