Near-zero feature extraction (single camera)

Yonim

Hi ModalAI team,

I'm running voxl-open-vins-server on a VOXL2 Mini (SDK 1.6.2) with two AR0144 tracking cameras in a non-overlapping front + down configuration (not stereo). I would appreciate guidance.

ENVIRONMENT

VOXL2 Mini, SDK 1.6.2
voxl-open-vins-server, internal tracker (en_ext_feature_tracker not used; voxl-feature-tracker disabled)
2x AR0144, front + down, non-overlapping FOV
estimator_config.yaml (fpv): use_stereo: false, max_cameras: 2, use_klt: true, use_gpu: true, num_pts: 50, fast_threshold: 15, min_px_dist: 10, histogram_method: HISTOGRAM
vio_cams.conf: both cams enabled

The issue is Near-zero feature extraction (single camera): I disabled the down camera. However, even during motion over textured, well-lit scenes, voxl-inspect-vins reports only ~1 feature, quality 0%, state OKAY, and T_imu_wrt_vio runs away (tens of meters). The estimator loop is healthy but the tracker extracts almost nothing.

What I've verified / ruled out:

Both camera feeds live and clean at the pipe level: 1280x800 @ 30fps, RAW8, reasonable exposure/gain (voxl-inspect-cam).
Intrinsics present and valid (fisheye, both cams); extrinsics validated incl. body -> imu_apps (voxl-inspect-extrinsics passes).
IMU vibration at rest all green (voxl-inspect-vibration).
Tried histogram_method HISTOGRAM and NONE; min_px_dist 50 -> 10; use_gpu true/false (CPU path "sticks"); switched tracking pipes off the _ion variants to plain *_misp_norm. None of these moved the feature count off ~1.

Questions:

For the single-camera case, what would cause the internal tracker to return ~1 feature on a sharp, textured image with the config above?
Is there a recommended way to visualize the internal tracker's extracted features for OpenVINS (the cameras-tab feat_overlay pipe is empty since voxl-feature-tracker is disabled — is ov_overlay under the VIO tab the correct view)?
Is there updated AR0144 driver / SDK guidance for non-overlapping multi-camera VIO on the VOXL2 Mini?

Best regards,
Yoni