Cause of Performance Saturation with Increasing Camera Count

Determine whether the observed decrease in crash rate and increase in success rate, followed by performance saturation as the number of depth cameras increases, are driven by the expansion of the combined field of view until additional sensors cease to provide novel or actionable environmental information.

Background

The navigation policy is evaluated with one through seven depth cameras, although configurations with five, six, and seven cameras are out of distribution because only up to four cameras were used during training. The reported results show improving navigation performance as camera count increases, with success rates eventually reaching a plateau.

The paper attributes this trend to increasing aggregate angular coverage, defined as the number of spherical-range-image bins covered by sensors divided by the total number of bins. However, the authors present this causal explanation only as a hypothesis and do not establish whether the performance trend and eventual plateau are in fact caused by the availability of novel or actionable visual information rather than other effects of sensor redundancy or policy behavior.

References

We hypothesize that this trend is driven by the expansion of the combined \ac{fov}; as more cameras are integrated, the aggregate angular coverage (number of bins covered by a sensor divided by total number of bins) increases until additional sensors no longer provide the policy with novel or actionable information about the surrounding environment.

— Sensor-Layout-Agnostic Navigation via Geometric Observation Canonicalization  (2610.08306 - Rehberg et al., 6 Oct 2026) in Section 5.2.1, “Performance Evaluation”