Why event vision

A frame-based camera integrates light over an exposure and produces images at a chosen rate. That is a useful representation for many perception systems. For a moving aircraft, some combinations of motion, exposure, and lighting can make blur, latency, and dynamic range difficult engineering trade-offs.

An event camera makes a different trade. Each pixel is independent and reports only when the brightness it sees changes, emitting a timestamp, an address and a polarity. Events are asynchronous rather than complete frames. Their rate depends on scene changes, sensor settings, and noise; a busy scene can produce a substantial processing load.

What that buys

Three properties make the sensing modality worth investigating, with actual performance depending on the sensor and operating conditions:

• Fine temporal resolution without waiting for the next full frame. Timestamp resolution is not the same as sensor or end-to-end processing latency.
• The opportunity for sparse computation when event activity is low, with overload and noise still needing explicit handling.
• Wide dynamic range in suitable conditions, without guaranteeing useful motion estimates in darkness, low texture, or every lighting transition.

Those are the reasons the sensor is interesting. They are not, on their own, a navigation system.

The harder half

Many established vision pipelines expect frames. Accumulating events into images can make integration easier, but trades away some of the timing information. Whether a native event pipeline is better is an empirical question.

So the question becomes: what computation is native to this data? One answer has been sitting in the neuroscience literature for decades. The fly solves visual navigation with a small set of specialised circuits that correlate neighbouring inputs over a short delay, producing a local estimate of motion directly, with no image in between. It is local, parallel, asynchronous, and it maps onto sparse event streams far more naturally than onto frames.

That correspondence is the bet haltere is built on: not that biology is elegant, but that this particular architecture fits this particular sensor.

Where we are

As of 5 September 2026: specification complete, implementation starting. Nothing has flown. The first scope is recorded-data motion perception and a customer-relevant comparison, not complete autonomous navigation.

The technical brief describes the proposed pipeline and its limitations. Gallego et al., Event-based Vision: A Survey provides background on the modality and its challenges; it is not evidence of Haltere's performance.