Search papers, labs, and topics across Lattice.
This study benchmarks various neural 3D reconstruction methods, including NeRF and 3D Gaussian Splatting, across different compute platforms to assess their real-time viability for autonomous laboratory robots. The findings reveal that while Gaussian Splatting achieves superior rendering quality compared to NeRF, it incurs a higher GPU cost, and onboard compute capabilities are inadequate for full per-scene optimization at interactive rates. Additionally, the evaluation of Meta's SAM3D shows it can generate plausible object geometry quickly, but with detail mismatches that may hinder effective manipulation, suggesting a need for tiered reconstruction pipelines in robotic applications.
Gaussian Splatting outperforms NeRF in rendering quality, but at a steep GPU cost that challenges real-time application in autonomous robots.
Autonomous robots performing laboratory tasks depend on 3D reconstruction pipelines that can turn raw camera streams into actionable object representations within the latency budget of a physical control loop. Neural 3D reconstruction methods have demonstrated high-quality view synthesis, but their real-time viability across the compute platforms on which laboratory robots actually run remains poorly characterized. In this work, we present a systematic compute-platform benchmark of neural 3D reconstruction methods, evaluating NeRF and 3D Gaussian Splatting training and rendering on GPU-enabled computing devices ranging from single-board computers to server-class nodes, and place Meta's SAM3D single-image reconstruction on the same axes to quantify its latency and fidelity gap relative to per-scene optimization. Our results show that Gaussian Splatting yields higher rendering quality than NeRF at greater GPU cost, and that onboard compute is insufficient for full per-scene optimization at interactive rates. Our preliminary assessment on SAM3D indicates that it delivers plausible object geometry within seconds, but with detail mismatches that can compromise downstream manipulation. Together, these findings motivate tiered pipelines in which lightweight feed-forward reconstruction sustains the real-time perception-and-tracking loop for laboratory robots, while heavier neural reconstruction is scheduled selectively on suitable compute.
