Answers are grounded in this paper and cite their sources.
Mobile robots pair solid-state LiDARs with cameras to acquire data, yet cross-stage error coupling in low-SWaP-C payloads remains understudied. We present staged calibration and validation of a mobile payload pairing a Livox Mid-360S LiDAR and FLIR stereo cameras on a Jetson AGX Orin (32 GB). An unmodeled 2.5 mm target bow induces 4.4% focal length bias (4.6% depth error), producing a 133 mm lateral phantom extrinsic translation downstream. Staged calibration contains errors via: (1) a laser scale gate bounding scale to −1.6±1.0%; (2) symmetric-pitch observability resolving vertical ill-conditioning (κ = 3.66); and (3) empirical count-squared weighting. Trees’ stem width depends on edge convention (−4.62%±1.27% to +5.77%±1.38%). Across three eastern white pine stands, closed-loop SLAM evaluation against 65 taped trees (15/17 located in Stand 1) yields single-slice diameter bias of +3.26±2.53 cm (RMSE 4.12 cm, R2 = 0.9510) across all stems, and 3.18 cm RMSE (+2.84 ± 1.46 cm, R2 = 0.9836) on full rings (≥ 340◦, n = 42). The bias is an additive radial offset (≈ +1.42 cm, independent of diameter) whose cause is not separated here, with an empirical coverage penalty (+2.79 cm/100◦ missing arc) and a fit-residual trap where partial arcs yield deceptively low residuals. On LiDAR-gated clear stems (n = 52), zero-shot SAM ViT-B recovers single-view diameter to +1.92 ± 3.44 cm (MAE 2.90 cm) versus LiDAR’s +3.38 ± 4.07 cm.
No takes yet. Share an insight, caveat, or question.
Garms et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: