New SLAM benchmark exposes where event cameras outrun their algorithms

Event-based SLAM Benchmark for High-Speed Maneuvers (arxiv 2604.24033) introduces a dataset and evaluation protocol specifically designed to stress-test event-camera SLAM algorithms under the conditions where frame-camera approaches fail completely. The results are instructive: the sensor hardware holds up well; the algorithms do not, at least not yet.

What the benchmark measures

Existing event-camera SLAM benchmarks were collected on handheld rigs moving at walking speed or on slow ground vehicles. They captured the advantages of event cameras at moderate motion rates — better temporal resolution than frames, cleaner output in difficult lighting — but they did not expose the failure modes that appear at high angular rates.

The new benchmark uses a purpose-built rig capable of angular rates above 1000 deg/s, which corresponds roughly to the motion environment on an airframe during an agile intercept maneuver. At these rates, several classes of algorithm that work well at moderate speeds fail to produce useful estimates.

Where the algorithms break

The survey identifies three breakdown modes under high-speed motion:

Event synchronization. At very high angular rates, the temporal correspondence between an event and the scene feature that generated it becomes ambiguous. Events from the leading edge of a feature alias with events from the trailing edge, and algorithms that assume feature-level correspondence break down.

Filter divergence. Kalman-filter-based estimators trained on moderate-motion data diverge under high-rate inputs because the motion model does not capture the dynamics correctly. Retuning is required for each platform and motion regime — a calibration burden that matters for a seeker that sees different dynamics at different points in an engagement.

Representation latency. Some event-camera SLAM algorithms convert asynchronous events into frame-like representations before processing. At high event rates this conversion introduces latency that cancels much of the advantage of using an event camera in the first place.

The positive finding

Despite the breakdown catalog, the benchmark reveals something important: the sensor does not fail at high angular rates. Events continue to arrive accurately and in the correct temporal order even at 1000 deg/s. The failures are algorithmic, not hardware limitations. This is a tractable problem.

The benchmark authors provide baseline results for six representative algorithms and make the dataset publicly available. For groups working on event-camera estimation — particularly for guidance systems — this gives a concrete target: match the baseline algorithms at moderate speed, then close the gap at high speed.

Relevance to terminal guidance

For a seeker approaching a maneuvering target, high angular rates are not edge cases — they are the operating regime in the last few seconds. An estimator that cannot maintain a valid track at 500 deg/s angular rate is not useful at intercept.

The benchmark does not provide answers to this problem, but it provides the right questions, and it provides them with reproducible measurements. The next generation of event-based SLAM and tracking algorithms will be validated against this benchmark, and the ones that pass will be candidates for embedded guidance systems.

---

Event-based SLAM Benchmark for High-Speed Maneuvers. arxiv 2604.24033