Pith. sign in

REVIEW 2 major objections 4 minor 36 references

SplatCtrl turns live RGB-D streams into continuous distance fields so a robot arm can reconstruct unknown scenes and replan collision-free 6-DoF motion in real time.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 05:33 UTC pith:4K6L75SG

load-bearing objection Competent systems paper that closes a real online loop from RGB-D Gaussians to reactive 6-DoF CBF control; the SDF approximation is the soft link but the empirical results still hold. the 2 major comments →

arxiv 2607.08948 v1 pith:4K6L75SG submitted 2026-07-09 cs.RO cs.CV

SplatCtrl: Perception-Action Coupling via Gaussian Scene Representations and Reactive Robot Control

Siddarth Jain , Ho Jin Choi This is my paper
classification cs.RO cs.CV
keywords 3D Gaussian Splattingreactive robot controlsigned distance functionscontrol barrier functionsperception-action couplingRGB-D scene reconstructioncollision avoidance
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Robots that work well on factory floors usually fail when objects move or appear without warning, because they lack a scene model that can be rebuilt and used for control at the same time. SplatCtrl claims that a fixed budget of isotropic 3D Gaussians, filtered by voxels and relocated on the fly, can serve as that model: it is updated from ordinary RGB-D cameras and simultaneously supplies a continuous, differentiable signed-distance field. That distance field is fed into control-barrier functions inside a quadratic-program inverse-kinematics controller, producing smooth, reactive arm motions that stay clear of both static clutter and moving people. The authors show that the same loop works in simulation, on a physical 7-DoF arm across twelve previously unseen tabletop scenes, and in a shared human-robot workspace, reaching roughly 96 percent success without offline training or pre-built maps. A sympathetic reader cares because the method removes the usual separation between perception and planning, letting a manipulator operate in the same open-ended spaces humans already occupy.

Core claim

A hybrid of voxel-filtered, dynamically relocated isotropic Gaussians and a mass-lumped Gaussian-process distance field yields a numerically stable continuous SDF that can be inserted directly into control-barrier functions, enabling the first reported real-time system that simultaneously reconstructs unknown dynamic scenes from RGB-D streams and generates full 6-DoF collision-free arm motion.

What carries the argument

The continuous SDF obtained from isotropic Gaussians via a mass-lumped Gaussian-process distance field (GPDF). It converts the discrete set of spheres into a smooth, differentiable occupancy and distance estimate whose gradient supplies the barrier constraint for the QP-IK controller.

Load-bearing premise

That treating every Gaussian as a simple sphere and then approximating the Gaussian-process matrix by its row sums still produces a distance field accurate enough for safe barrier-function control.

What would settle it

Run the identical physical-robot trials while replacing the isotropic-sphere GPDF with either anisotropic full-covariance Gaussians or a ground-truth mesh SDF; if collision rates or minimum clearance systematically worsen under the paper's method, the central claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. SplatCtrl presents a unified real-time framework that reconstructs dynamic scenes from multi-view RGB-D streams via isotropic 3D Gaussian Splatting (with hybrid voxel filtering and dynamic relocation) and couples the resulting representation to reactive 6-DoF robot control. Continuous signed-distance and collision-probability fields are obtained from the Gaussians by a mass-lumped Gaussian Process Distance Field (GPDF) approximation (Eqs. 6–16); these fields supply differentiable barrier constraints inside a QP inverse-kinematics controller (Eq. 17). The system is evaluated in simulation (942 trials across three environments with camera-count ablations), on a physical Franka arm (24 trials in twelve unseen scenes), and in a pilot human–robot shared-workspace study (n=8), reporting high success rates and safer, faster completion relative to a safety-monitored-stop baseline.

Significance. If the continuous GPDF-to-CBF bridge is sufficiently accurate and conservative, the work supplies a practical, training-free perception–action loop that lets manipulators operate in previously unknown, continuously changing workspaces—an important step beyond static maps or offline Gaussian pipelines. The explicit numerical-stability rewrite (Eq. 12), mass-lumping reduction to linear complexity, and dual-pipeline (BiRRT + QP-IK) real-robot validation with multi-view RGB-D are concrete engineering contributions. Concurrent Gaussian navigation systems exist, yet the manipulator-specific 6-DoF reactive formulation and the reported success rates under online reconstruction remain valuable for the robotics community.

major comments (2)
  1. [§III.D, Eqs. 6–16; §IV.C] §III.D (Eqs. 6–16) and the Discussion state that the proxy SDF “typically underestimates true surface distance,” that isotropic spheres “overlook their full probabilistic characteristics,” and that sphere marching is optional. No quantitative residual-error distribution of d_SDF versus ground-truth surface distance (or false-negative rate of the barrier) is reported under the fixed 5 k Gaussian budget and multi-view noise used in the real-robot trials. Because the CBF constraint in Eq. 17 relies on this distance for safety, the underestimation could render the barrier non-conservative precisely when the central claim of reliable collision-free 6-DoF control is asserted; a short error analysis or measured clearance histogram would close the gap.
  2. [Table II; §IV.C; Abstract] Table II and §IV.C report point success rates (95 % simulation, 95.8 % real) without error bars, confidence intervals, or per-trial minimum-distance statistics beyond the qualitative statement that the 2 cm threshold was “largely satisfied.” Given that BiRRT occasionally violated the margin and one QP-IK trial passed close, the absence of residual-risk quantification weakens the safety claim that underpins the “first real-time full 6-DoF” contribution listed in the Abstract and §I.
minor comments (4)
  1. [Throughout] Several typographic artifacts appear throughout (e.g., “V oxel”, “ManiGaus-sian”, “Mat ´ern”, “G P”, “ˆo(x)”). A careful proof-reading pass would improve readability.
  2. [Fig. 6] Figure 6 captions and axes are not fully self-contained; units and the precise definition of “joint length” / “EEF length” should be stated in the caption or legend.
  3. [§IV.D] The human–robot study (n=8) is correctly labeled a pilot, yet the text occasionally generalizes its findings; a single clarifying sentence that larger-scale validation remains future work would avoid over-statement.
  4. [§III.C–D] Hyper-parameter values (α_min, occupancy cut-offs 0.3/0.7, variance scale factor 9, l=2 cm) are listed but their sensitivity is not discussed; a brief ablation or justification would help reproducibility.

Circularity Check

0 steps flagged

No significant circularity: constructive RGB-D-to-Gaussian-to-GPDF-SDF-to-CBF pipeline validated against external collision/goal metrics.

full rationale

The derivation chain is self-contained and non-circular. Isotropic Gaussians are fitted to RGB-D observations via standard photometric/depth losses (Eq. 4) plus voxel filtering and relocation heuristics; the continuous SDF is then obtained by extending an external GPDF construction (Eqs. 6–16, citing [24]) with mass-lumping and optional sphere marching; these distances enter CBF constraints inside a standard QP-IK (Eq. 17). Success rates (simulation Table II, real-robot 95.8 %, human-robot study) are measured by independent criteria—goal pose error, actual collisions, and separation distances—not by quantities that were fitted to produce those numbers. Hyper-parameters (Gaussian budget, α_min, l, voxel size) are chosen by hand but do not render the reported outcomes tautological. Self-citations are absent from the load-bearing steps; prior Gaussian-splatting and CBF literature is external. Minor residual risk exists in the acknowledged underestimation of the proxy SDF, but that is a correctness/approximation issue, not circularity. Score 1 reflects only the ordinary presence of hand-tuned constants, not any definitional or fitted-input loop.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 2 invented entities

The central claim rests on standard robotics and Gaussian-process tools plus a handful of engineering choices (isotropic spheres, mass lumping, fixed Gaussian budget, hand-tuned thresholds). No new physical entities are postulated; the free parameters are ordinary hyper-parameters of the reconstruction and control stack.

free parameters (6)
  • Gaussian budget N_max
    Hard limit on number of Gaussians (5 000 in real experiments) chosen for real-time performance; directly affects reconstruction fidelity and collision accuracy.
  • opacity thresholds α_init=0.5, α_min=0.1
    Hand-set values that decide which Gaussians survive filtering and therefore which surfaces enter the SDF.
  • GPDF length-scale l (2 cm)
    Interpolation length that controls smoothness of the continuous SDF; chosen by hand and used for both simulation and real-robot trials.
  • loss weights λ=0.2, λ_color=λ_depth=1, λ_reg=0.02
    Balancing terms in the Gaussian optimization objective; affect final placement of Gaussians.
  • occupancy free/uncertain cutoffs 0.3 / 0.7
    Voxel thresholds that decide which Gaussians are culled as free-space artifacts.
  • safety margin ε and variance scale factor 9
    Barrier offset and ad-hoc variance division that keep collision probability near zero far from surfaces.
axioms (4)
  • domain assumption Isotropic covariance Σ_i = r_i² I is an adequate geometric model for collision geometry of real objects.
    Stated in §III.B; later acknowledged as a limitation in the Discussion.
  • ad hoc to paper Mass-lumped diagonal approximation of the GP kernel matrix preserves a usable SDF and positivity of occupancy.
    Introduced in §III.D.3 to reduce complexity from O(n³) to linear; correctness of the resulting barrier is assumed rather than proved.
  • domain assumption Control-barrier functions formulated with the proxy SDF guarantee forward invariance of the safe set under the QP-IK dynamics.
    Standard CBF theory is invoked; the paper does not re-derive the invariance proof for the approximate GPDF field.
  • domain assumption RGB-D streams from a small number of fixed cameras plus robot kinematics are sufficient to keep the Gaussian map consistent with the true workspace.
    Implicit throughout the experimental sections; multi-view ablation shows sensitivity to camera count.
invented entities (2)
  • Hybrid voxel-filtered dynamic Gaussian relocation strategy no independent evidence
    purpose: Maintain a compact, artifact-free isotropic Gaussian map under continuous environmental change.
    Engineering construct assembled from existing 3D-GS densification ideas plus occupancy grids; no independent physical existence claimed.
  • Continuous SDF / collision-probability field derived from isotropic Gaussians via mass-lumped GPDF no independent evidence
    purpose: Supply differentiable distance and gradient signals for control-barrier constraints.
    Mathematical construction that extends prior GPDF work to spheres; falsifiable only through the reported collision-avoidance success rates.

pith-pipeline@v1.1.0-grok45 · 17062 in / 3258 out tokens · 31109 ms · 2026-07-13T05:33:56.700574+00:00 · methodology

0 comments
read the original abstract

Robotic manipulators excel in structured environments but face substantial challenges in unstructured and dynamic settings. This paper presents SplatCtrl, a unified framework for real-time scene reconstruction and reactive robot motion generation to enable collision-free robotic arm control in previously unseen and continuously changing environments. Building on 3D Gaussian Splatting (3D-GS), we introduce a hybrid voxel-based filtering and dynamic Gaussian relocation strategy that supports efficient scene reconstruction from RGB-D streams while accommodating environmental changes. For safe and reactive control, we further propose a method for deriving continuous signed distance functions from isotropic Gaussians, providing stable and differentiable collision probability estimates that bridge classical distance fields with the modern implicit representation. These continuous distance metrics are incorporated into control barrier functions, resulting in a unified perception-action coupling framework that supports smooth and reliable real-time motion generation in response to scene changes. Experimental validation in simulation, on physical robot, and within shared human-robot workspace demonstrates the framework's effectiveness, achieving integrated scene reconstruction and reactive control in uncertain, and dynamic environments.

Figures

Figures reproduced from arXiv: 2607.08948 by Ho Jin Choi, Siddarth Jain.

Figure 1
Figure 1. Figure 1: SplatCtrl leverages real-time RGB-D data to re [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SplatCtrl, including scene reconstruction via isotropic Gaussian splatting, occupancy-based density [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The qualitative comparison of our method with 3D [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Left and Middle: Illustrations of the Cubby and Tabletop simulation environments [35]. Right: Experimental setup for real-robot tasks. Three Intel RealSense D435 cam￾eras (Cameras 1–3) provide multi-view RGB-D observations for scene reconstruction. For the human–robot study, an additional Intel L515 camera (Camera 4) is included. B. Simulation Robot Experiments Reconstruction quality primarily measures sce… view at source ↗
Figure 5
Figure 5. Figure 5: Evaluation in twelve unseen environments containing obstacles that vary in geometry, layout, color, height, and [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of perception–action coupling using [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Sequential keyframes from the human–robot shared workspace trial. The robot transports a yellow cube from the left [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Compared to the Safety Monitored Stopping (SMS) [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 13 linked inside Pith

  1. [1]

    Octomap: an efficient probabilistic 3d mapping framework based on octrees,

    A. Hornung, K. M. Wurm, M. Bennewitz, C. Stachniss, and W. Bur- gard, “Octomap: an efficient probabilistic 3d mapping framework based on octrees,”Autonomous Robots, vol. 34, pp. 189 – 206, 2013

    2013

  2. [2]

    Pointnet: Deep learning on point sets for 3d classification and segmentation,

    C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “Pointnet: Deep learning on point sets for 3d classification and segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 652–660

    2017

  3. [3]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoor- thi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Comm. of the ACM, vol. 65, pp. 99 – 106, 2020

    2020

  4. [4]

    3d gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering,” inACM Transactions on Graphics (SIGGRAPH), vol. 42, no. 4, 2023, pp. 1–14

    2023

  5. [5]

    Real- time perception meets reactive motion generation,

    D. Kappler, F. Meier, J. Issac, J. Mainprice, C. G. Cifuentes, M. W ¨uthrich, V . Berenz, S. Schaal, N. Ratliff, and J. Bohg, “Real- time perception meets reactive motion generation,”IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 1864–1871, 2018

    2018

  6. [6]

    Earl: Eye-on-hand reinforcement learner for dynamic grasping with active pose estimation,

    B. Huang, J. Yu, and S. Jain, “Earl: Eye-on-hand reinforcement learner for dynamic grasping with active pose estimation,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2023, pp. 2963–2970

    2023

  7. [7]

    Neural mp: A generalist neural motion planner,

    M. Dalal, J. Yang, R. Mendonca, Y . Khaky, R. Salakhutdinov, and D. Pathak, “Neural mp: A generalist neural motion planner,”arXiv preprint arXiv:2409.05864, 2024

    Pith/arXiv arXiv 2024

  8. [8]

    Point cloud matters: Rethinking the impact of different observation spaces on robot learning,

    H. Zhu, Y . Wang, D. Huang, W. Ye, W. Ouyang, and T. He, “Point cloud matters: Rethinking the impact of different observation spaces on robot learning,”Advances in Neural Information Processing Systems, vol. 37, pp. 77 799–77 830, 2024

    2024

  9. [9]

    V oxelnet: End-to-end learning for point cloud based 3d object detection,

    Y . Zhou and O. Tuzel, “V oxelnet: End-to-end learning for point cloud based 3d object detection,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4490–4499

    2018

  10. [10]

    V olumetric grasping network: Real-time 6 dof grasp detection in clutter,

    M. Breyer, J. J. Chung, L. Ott, R. Siegwart, and J. Nieto, “V olumetric grasping network: Real-time 6 dof grasp detection in clutter,” in Conference on Robot Learning. PMLR, 2021, pp. 1602–1611

    2021

  11. [11]

    Nerfs in robotics: A survey,

    G. Wang, L. Pan, S. Peng, S. Liu, C. Xu, Y . Miao, W. Zhan, M. Tomizuka, M. Pollefeys, and H. Wang, “Nerfs in robotics: A survey,”The International Journal of Robotics Research, p. 02783649251374246, 2024

    2024

  12. [12]

    3d gaussian splatting in robotics: A survey,

    S. Zhu, G. Wang, X. Kong, D. Kong, and H. Wang, “3d gaussian splatting in robotics: A survey,”arXiv preprint arXiv:2410.12262, 2024

    Pith/arXiv arXiv 2024

  13. [13]

    Manigaussian: Dynamic gaussian splatting for multi-task robotic manipulation,

    G. Lu, S. Zhang, Z. Wang, C. Liu, J. Lu, and Y . Tang, “Manigaussian: Dynamic gaussian splatting for multi-task robotic manipulation,” in European Conference on Computer Vision, 2024

    2024

  14. [14]

    Splat-mover: Multi-stage, open- vocabulary robotic manipulation via editable gaussian splatting,

    O. Shorinwa, J. Tucker, A. Smith, A. Swann, T. Chen, R. Firoozi, M. Kennedy III, and M. Schwager, “Splat-mover: Multi-stage, open- vocabulary robotic manipulation via editable gaussian splatting,”arXiv preprint arXiv:2405.04378, 2024

    Pith/arXiv arXiv 2024

  15. [15]

    Robo-gs: A physics consistent spatial-temporal model for robotic arm with hybrid representation,

    H. Lou, Y . Liu, Y . Pan, Y . Geng, J. Chen, W. Ma, C. Li, L. Wang, H. Feng, L. Shiet al., “Robo-gs: A physics consistent spatial-temporal model for robotic arm with hybrid representation,”arXiv preprint arXiv:2408.14873, 2024

    Pith/arXiv arXiv 2024

  16. [16]

    Splatsim: Zero-shot sim2real transfer of rgb manipulation policies using gaussian splatting,

    M. N. Qureshi, S. Garg, F. Yandun, D. Held, G. Kantor, and A. Silwal, “Splatsim: Zero-shot sim2real transfer of rgb manipulation policies using gaussian splatting,”arXiv preprint arXiv:2409.10161, 2024

    Pith/arXiv arXiv 2024

  17. [17]

    Let us make a splan: Risk-aware trajectory optimization in a normalized gaussian splat,

    J. Michaux, S. Isaacson, C. E. Adu, A. Li, R. K. Swayampakula, P. Ewen, S. Rice, K. A. Skinner, and R. Vasudevan, “Let us make a splan: Risk-aware trajectory optimization in a normalized gaussian splat,”IEEE Transactions on Robotics, vol. 41, pp. 4380–4397, 2025

    2025

  18. [18]

    Foci: Trajectory optimization on gaussian splats,

    M. G. Andreu, M. Wilder-Smith, V . Klemm, V . Patil, J. Tordesillas, and M. Hutter, “Foci: Trajectory optimization on gaussian splats,” ArXiv, vol. abs/2505.08510, 2025. [Online]. Available: https: //api.semanticscholar.org/CorpusID:278535390

    Pith/arXiv arXiv 2025

  19. [19]

    Splat-nav: Safe real-time robot navigation in gaussian splatting maps,

    T. Chen, O. Shorinwa, W. Zeng, J. Bruno, P. Dames, and M. Schwa- ger, “Splat-nav: Safe real-time robot navigation in gaussian splatting maps,”ArXiv, vol. abs/2403.02751, 2024

    Pith/arXiv arXiv 2024

  20. [20]

    Safer-splat: A control barrier function for safe navigation with online gaussian splatting maps,

    T. Chen, A. Swann, J. Yu, O. Shorinwa, R. Murai, M. Kennedy III, and M. Schwager, “Safer-splat: A control barrier function for safe navigation with online gaussian splatting maps,”arXiv preprint arXiv:2409.09868, 2024

    Pith/arXiv arXiv 2024

  21. [21]

    Splatam: Splat, track & map 3d gaussians for dense rgb-d slam,

    N. V . Keetha, J. Karhade, K. M. Jatavallabhula, G. Yang, S. Scherer, D. Ramanan, and J. Luiten, “Splatam: Splat, track & map 3d gaussians for dense rgb-d slam,”ArXiv, vol. abs/2312.02126, 2023

    Pith/arXiv arXiv 2023

  22. [22]

    Isotropic gaussian splatting for real-time radiance field rendering,

    Y . Gong, L. Yu, and G. Yue, “Isotropic gaussian splatting for real-time radiance field rendering,”ArXiv, vol. abs/2403.14244, 2024

    Pith/arXiv arXiv 2024

  23. [23]

    Gs-planner: A gaussian-splatting-based planning framework for active high-fidelity reconstruction,

    R. Jin, Y . Gao, Y . Wang, H. Lu, and F. Gao, “Gs-planner: A gaussian-splatting-based planning framework for active high-fidelity reconstruction,”ArXiv, vol. abs/2405.10142, 2024

    Pith/arXiv arXiv 2024

  24. [24]

    Accurate gaussian-process-based distance fields with applications to echolocation and mapping,

    C. L. Gentil, O.-L. Ouabi, L. Wu, C. Pradalier, and T. Vidal-Calleja, “Accurate gaussian-process-based distance fields with applications to echolocation and mapping,”IEEE Robotics and Automation Letters, vol. 9, pp. 1365–1372, 2023

    2023

  25. [25]

    Towards feasible dynamic grasping: Leveraging gaussian process distance fields, se(3) equivariance and riemannian mixture models,

    H. J. Choi and N. Figueroa, “Towards feasible dynamic grasping: Leveraging gaussian process distance fields, se(3) equivariance and riemannian mixture models,” inIEEE International Conference on Robotics and Automation (ICRA), 2024

    2024

  26. [26]

    Distance and collision probability estimation from gaussian surface models,

    K. Goel and W. Tabib, “Distance and collision probability estimation from gaussian surface models,”arXiv:2402.00186, 2024

    Pith/arXiv arXiv 2024

  27. [27]

    Stochastic poisson surface reconstruction,

    S. Sellan and A. Jacobson, “Stochastic poisson surface reconstruction,” ACM Transactions on Graphics (TOG), vol. 41, pp. 1 – 12, 2022

    2022

  28. [28]

    A unified framework for coordinated multi-arm motion planning,

    S. S. M. Salehian, N. Figueroa, and A. Billard, “A unified framework for coordinated multi-arm motion planning,”The International Journal of Robotics Research, vol. 37, pp. 1205 – 1232, 2018

    2018

  29. [29]

    Constrained passive interaction control: Leveraging passivity and safety for robot manipulators,

    Z. Zhang, T. Li, and N. Figueroa, “Constrained passive interaction control: Leveraging passivity and safety for robot manipulators,”IEEE International Conference on Robotics and Automation (ICRA), 2024

    2024

  30. [30]

    Curobo: Parallelized collision-free robot motion generation,

    B. Sundaralingam, S. K. S. Hari, A. Fishman, C. R. Garrett, K. V . Wyk, V . Blukis, A. Millane, H. Oleynikova, A. Handa, F. T. Ramos, N. D. Ratliff, and D. Fox, “Curobo: Parallelized collision-free robot motion generation,”2023 IEEE International Conference on Robotics and Automation (ICRA), pp. 8112–8119, 2023

    2023

  31. [31]

    CVXPY: A Python-embedded modeling language for convex optimization,

    S. Diamond and S. Boyd, “CVXPY: A Python-embedded modeling language for convex optimization,”Journal of Machine Learning Research, vol. 17, no. 83, pp. 1–5, 2016

    2016

  32. [32]

    Large scale multi-view stereopsis evaluation,

    R. Jensen, A. Dahl, G. V ogiatzis, E. Tola, and H. Aanæs, “Large scale multi-view stereopsis evaluation,” in2014 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2014, pp. 406–413

    2014

  33. [33]

    Mvsnet: Depth inference for unstructured multi-view stereo,

    Y . Yao, Z. Luo, S. Li, T. Fang, and L. Quan, “Mvsnet: Depth inference for unstructured multi-view stereo,”European Conference on Computer Vision (ECCV), 2018

    2018

  34. [34]

    3d gaussian splatting as markov chain monte carlo,

    S. Kheradmand, D. Rebain, G. Sharma, W. Sun, J. Tseng, H. Isack, A. Kar, A. Tagliasacchi, and K. M. Yi, “3d gaussian splatting as markov chain monte carlo,”ArXiv, vol. abs/2404.09591, 2024

    Pith/arXiv arXiv 2024

  35. [35]

    Motion policy networks,

    A. Fishman, A. Murali, C. Eppner, B. N. Peele, B. Boots, and D. Fox, “Motion policy networks,” inConference on Robot Learning, 2022

    2022

  36. [36]

    Densepose: Dense human pose estimation in the wild,

    R. A. G ¨uler, N. Neverova, and I. Kokkinos, “Densepose: Dense human pose estimation in the wild,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7297–7306

    2018