Pith. sign in

REVIEW 4 major objections 5 minor 9 cited by

DISCOVERSE: Efficient Robot Simulation in Complex High-Fidelity Environments

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Combining Gaussian-splatting rendering with MuJoCo physics trains manipulation policies that transfer zero-shot to real robots at up to 86.5 percent average success.

desk verdict Solid systems paper with plausible zero-shot sim2real results, but the causal claim about visual fidelity is not yet separated from physics tuning and the benchmark lacks statistical rigor. read the letter →

arxiv 2507.21981 v1 pith:YWFFQIQX submitted 2025-07-29 cs.RO

Yufei Jia , Guangyu Wang , Yuhang Dong , Junzhe Wu , Yupei Zeng , Haonan Lin , Zifan Wang , Haizhou Ge
show 12 more authors
Weibin Gu Kairui Ding Zike Yan Yunjie Cheng Yue Li Ziming Wang Chuxuan Li Wei Sui Lu Shi Guanzhong Tian Ruqi Huang Guyue Zhou
This is my paper
classification cs.RO
keywords robotsimulationGaussiansplattingSim2RealtransferMuJoCoimitationlearningReal2Simphotorealisticrenderingmanipulation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Discoverse is a robot simulator that pairs 3D Gaussian Splatting rendering with the MuJoCo physics engine, with the aim of making simulation look close enough to reality that policies trained in it work on real robots. The paper's central claim is that this combination yields state-of-the-art zero-shot Sim2Real transfer on contact-rich manipulation. Averaged over closing a laptop, pushing a mouse, and picking up a kiwifruit, ACT policies trained only in Discoverse succeed 55 percent of the time without augmentation and 86.5 percent with image augmentation, compared with 44 and 68 percent for the best competing 3DGS simulator. The same pattern holds for Diffusion Policy: 56 and 86 percent versus 45.3 and 74.6 percent. If the claim holds, photorealistic radiance-field rendering plus rigid-body physics is a practical path to large-scale robot learning without collecting tens of thousands of human demonstrations.

What carries the argument

The central mechanism is the pairing of a 3D Gaussian Splatting renderer with a MuJoCo rigid-body simulator, connected by a dual 3DGS-Mesh asset format: Gaussian primitives produce photorealistic RGB and depth, while convex-decomposed meshes described in MJCF provide contact physics. A Mesh2GS transfer step initializes one flattened Gaussian per mesh facet, aligned to face normals, so the same object can be rendered as a radiance field and simulated as a rigid body. The Real2Sim pipeline also uses laser scanning, generative 3D reconstruction, estimated HDR environment lighting, and PBR as a preprocessing step to align object appearance with the reconstructed background.

What would settle it

Run the ACT and Diffusion Policy benchmarks in Discoverse twice: once with the 3DGS renderer and once with a conventional textured-mesh renderer, keeping the MuJoCo physics identical. If the success rates stay close to the reported values, photorealistic rendering is not the cause of the transfer gains and the paper's central claim fails; if they fall to the MuJoCo baseline, the rendering is confirmed as the load-bearing factor.

Watch

Extended reading notes

Core claim

Discoverse is presented as the first unified, modular, open-source 3DGS-based simulation framework for Real2Sim2Real robot learning. It reconstructs real scenes as Gaussian radiance fields, uses laser scanning, generative 3D models, and HDR relighting to build interactive objects, and converts those objects into a dual representation: 3DGS for rendering and convex-decomposed meshes in MJCF for MuJoCo physics. The experimental discovery is that imitation-learning policies trained in this simulator transfer zero-shot to the real world more reliably than policies trained in MuJoCo, RoboTwin, or SplatSim, with the advantage attributed to the photorealistic appearance that reduces the visual domain gap.

Load-bearing premise

The paper assumes MuJoCo's rigid-body contact model, with meshes split into convex pieces and friction, damping, and density set by hand within unspecified ranges, behaves like the real laptop, mouse, and kiwifruit closely enough that the measured Sim2Real gains come from visual fidelity rather than physics tuned to be easy.

Editorial extensions

If this is right

  • ACT policies trained in Discoverse transfer to the real world at 55 percent average success without augmentation and 86.5 percent with image augmentation, versus 44 and 68 percent for SplatSim.
  • Diffusion Policy follows the same ordering: 56 and 86 percent for Discoverse versus 45.3 and 74.6 percent for SplatSim.
  • Image-based augmentation closes much of the remaining visual gap, raising Discoverse's averages by 31.5 points for ACT and 29.3 points for Diffusion Policy.
  • Demonstration collection becomes roughly 100 times faster: 100 demonstrations take about 1.5 minutes in simulation against 146 minutes of human teleoperation in the real world.
  • Because the same framework supports RGB, depth, LiDAR, tactile sensing, and multiple robot embodiments, the benchmark can extend beyond tabletop manipulation to navigation and multi-agent coordination.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A consequence the authors leave implicit: if visual fidelity is the main carrier of the gain, 3DGS rendering could be bolted onto any physics engine as a modular upgrade, so the simulator's full-stack design may matter less than its renderer.
  • A blinded sweep over friction, damping, and density values would separate the renderer's contribution from the hand-tuned contact model; the paper does not report such a sweep, so the attribution to visual fidelity is not yet isolated.
  • The real-robot baselines sit near 94 to 100 percent, so a measurable Sim2Real gap remains; testing whether generative randomization fully closes it is a natural next step.
  • The three benchmark objects are rigid and roughly Lambertian; extending to deformable, transparent, or highly specular objects would stress both the 3DGS reconstruction and MuJoCo's contact model.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces DISCOVERSE, a unified open-source robot simulation framework that combines 3D Gaussian Splatting rendering with the MuJoCo physics engine and a ROS2 interface. It proposes a Real2Sim pipeline for reconstructing high-fidelity background scenes and interactive objects, plus domain-randomization mechanisms. The central claim is that policies (ACT and Diffusion Policy) trained in DISCOVERSE transfer zero-shot to the real world with significantly higher success rates than policies trained in MuJoCo, RoboTwin, or SplatSim, across three contact-rich manipulation tasks. The paper also reports high rendering throughput, automated demonstration collection, and exemplary navigation and multi-agent applications.

Significance. If the reported results hold, DISCOVERSE is a substantive engineering contribution: it is open-source, modular, supports multiple sensor modalities and robot embodiments, and offers an integrated Real2Sim pipeline that combines scanning, generation, relighting, and mesh-to-Gaussian transfer. The benchmark comparing four simulators under two imitation-learning policies and three real-world tasks is useful, and the reported throughput and automated data collection are practical assets. However, the paper's headline causal claim — that the Sim2Real gains are due to visual fidelity — is not yet isolated from the physics engine choice and manually adjusted physical parameters. The comparison is also missing two directly related 3DGS-based simulators (RoboGSim and Robo-GS) and lacks statistical rigor. The engineering value is real, but the evidence as presented is not sufficient to support the strong state-of-the-art and fidelity-attribution claims.

major comments (4)
  1. [§III-B Asset Formatting] This section states that physical properties for interactive objects (friction, damping, density) are 'randomized or manually adjusted ... within appropriate ranges,' but no values, ranges, or calibration procedures are reported. The benchmark in Tables II and III compares DISCOVERSE (MuJoCo physics with these tuned properties) against SplatSim (Bullet physics) and a MuJoCo baseline, so the 11–18 percentage-point margins are a joint test of rendering fidelity, physics engine, and manual tuning. The conclusion that the gains come from 'superior fidelity' is therefore underdetermined. Please report the exact physical parameters used, calibrate them against real object dynamics (e.g., friction and mass measurements), and include a controlled ablation that isolates rendering fidelity from physics tuning — for example, use identical mesh and physics parameters across renderers, or add a SplatSim-with-MuJoCo condition.
  2. [§V-A Evaluation Protocol and Tables II–III] Each task reports 50 test trials, but there are no confidence intervals, seeds, number of training runs, or statistical significance tests. Given that the margins are 11–18 percentage points and the underlying policy training is stochastic, the 'state-of-the-art' claim needs interval estimates (e.g., Wilson intervals across trials) and ideally multiple training seeds. Without this, the reader cannot assess whether the reported differences are reliable. Please add seed information, confidence intervals, and an appropriate test or at least show per-seed variability.
  3. [§II-B and Table I] RoboGSim [16] and Robo-GS [17] are cited and described as 3DGS-based Real2Sim simulators, yet neither appears in Table I or in the benchmark. Because the paper claims to be the 'first unified ... 3DGS-based simulation framework' and the 'state-of-the-art' simulator, omitting the most closely related systems from the comparison is a significant gap. Please include these baselines if feasible, or provide a concrete justification based on unavailability or incompatibility. At minimum, the 'first' claim should be qualified as 'to our knowledge' and the comparison set explicitly stated.
  4. [§VI Conclusion and Future Work] The paper's own future-work paragraph says 'advanced physical simulation' remains to be improved, which aligns with the concern that the current physics model is not calibrated. This admission, combined with the manual tuning in §III-B, means the causal statements in the Introduction (that Sim2Real gaps 'primarily originate from visual discrepancies' and that DISCOVERSE's gains are due to visual fidelity) go beyond the evidence. Please reframe the claims as demonstrating the joint Sim2Real performance of the full DISCOVERSE system, and avoid attributing the improvement to rendering alone until the physics confound is addressed.
minor comments (5)
  1. [§VI Heading] The heading 'CONCLUSION AND FUTHER WORK' contains a typo; it should be 'FURTHER WORK.'
  2. [Table I] The row for DISCOVERSE renders '✓ ⋆ ⋆ ⋆' rather than using the same star notation as other rows; this makes the table difficult to parse and should be cleaned up.
  3. [§IV-B Mesh-Gaussian Transfer] The phrase 'tagent axes' is a typo for 'tangent axes.'
  4. [§V-A Evaluation Protocol] The hardware components (AIRBOT-Gripper-2, LRCP V1080P cameras) and the 'gamepad-based state generation' procedure are not described in enough detail for reproduction; please add brief specifications or a reference to a documentation page.
  5. [References] Reference [43] is the GPT-4V system card and is cited in the context of a domain-randomization pipeline; please cite the specific technical report in a consistent format, as the current entry is incomplete.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central Sim2Real claims rest on externally measured robot trials, not on fitted parameters or self-cited constraints.

full rationale

DISCOVERSE is a systems and benchmarking paper; it makes no analytic derivation whose output is equivalent to its input. The headline claim is the zero-shot Sim2Real success rate, measured by deploying ACT and Diffusion Policy policies on a physical AIRBOT Play arm after training in each simulator. These are external, out-of-distribution measurements, not numbers reconstructed from a fitted model, so the comparison is not circular by construction. The only citation with overlapping authorship is ActiveSplat [23], which appears solely in an illustrative navigation application ('The agent takes ego-view renderings as input and progressively updates the spatial map [23]'), not in the benchmark tables or in any load-bearing argument. The manual adjustment of physical properties in Sec. III-B ('We either randomize or manually adjust physical properties, such as friction, damping, and density, within appropriate ranges') is a legitimate confound for the causal claim that visual fidelity drives the gain, because Discoverse's contact dynamics may have been tuned favorably. However, this is a validity threat, not circularity: no success rate is fitted to the simulation parameters, the physics engine is shared with the MuJoCo baseline, and the real-robot deployment is an independent arbiter. No equation, definitional identity, or self-citation forces the reported result, so the paper is self-contained with respect to circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on domain assumptions about rendering and physics fidelity, and on hand-chosen ranges for randomization and physical parameters. None are fitted to the target results in a circular way, but they are under-specified and not independently verified.

free parameters (2)
  • Domain randomization ranges (video overlay blend, HSV, gamma) = not specified
    Sec. IV-C applies these heuristics; the reported +31.5% and +29.3% improvements from augmentation depend on these hand-chosen ranges.
  • Physical property ranges (friction, damping, density) for interactive objects = not specified
    Sec. III-B says properties are randomized or manually adjusted within appropriate ranges, but no values are given; physics fidelity of the benchmark depends on them.
assumptions (4)
  • domain assumption 3D Gaussian Splatting is an adequate visual representation for training robot policies that transfer to reality.
    The whole framework renders with 3DGS; if splatting artifacts break policy features, the Sim2Real gains would not materialize. Invoked in Sec. III-A and Sec. IV.
  • domain assumption MuJoCo's rigid-body contact model with convex-decomposed meshes captures the real contact dynamics of the benchmark objects.
    Sec. III-B states meshes are decomposed into convex parts and physical properties are randomized or manually adjusted; contact-rich success rates depend on this.
  • domain assumption Laser-scanned and generative-model reconstructions are accurate enough to serve as the test environment for zero-shot transfer.
    Sec. IV-A and IV-B rely on LixelKity K1, Artec Leo, CLAY2, and DiffusionLight; reconstruction errors would confound the transfer results.
  • domain assumption Three tabletop tasks with 50 trials per condition are representative of the claimed state-of-the-art Sim2Real performance.
    Sec. V-A describes the benchmark; no power analysis or variance reporting supports generalization to other tasks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DISCOVERSE: Efficient Robot Simulation in Complex High-Fidelity Environments." pith.science (2026). https://pith.science/paper/YWFFQIQX

@misc{pith2026250721981,
  author       = {Pith},
  title        = {Pith review of: DISCOVERSE: Efficient Robot Simulation in Complex High-Fidelity Environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YWFFQIQX}},
  note         = {Machine review of arXiv:2507.21981}
}
read the original abstract

We present the first unified, modular, open-source 3DGS-based simulation framework for Real2Sim2Real robot learning. It features a holistic Real2Sim pipeline that synthesizes hyper-realistic geometry and appearance of complex real-world scenarios, paving the way for analyzing and bridging the Sim2Real gap. Powered by Gaussian Splatting and MuJoCo, Discoverse enables massively parallel simulation of multiple sensor modalities and accurate physics, with inclusive supports for existing 3D assets, robot models, and ROS plugins, empowering large-scale robot learning and complex robotic benchmarks. Through extensive experiments on imitation learning, Discoverse demonstrates state-of-the-art zero-shot Sim2Real transfer performance compared to existing simulators. For code and demos: https://air-discoverse.github.io/.

Figures

Figures reproduced from arXiv: 2507.21981 by the authors.

Figure 1
Figure 1. DISCOVERSE system overview. DISCOVERSE unifies real-world captures, 3D AIGC, and any existing 3D assets in formats of 3DGS (.ply), mesh (.obj/.stl), and MJCF physical models (.xml), enabling their use as interactive scene nodes (objects and robots) or the background node. We leverage Gaussian splatting as our rendering engine to generate hyper-realistic radiance field rendering of multiple sensor modalities and use … view at source ↗
Figure 2
Figure 2. DISCOVERSE operation flow. We utilize fast tile-based splatting for high-fidelity neural rendering and integrate MuJoCo [1] physical simulator for various robotic utilities. A. Engine We adopt the tile-based renderer in 3D Gaussian Splat￾ting [19] to simulate high-fidelity visuals and use the open￾source MuJoCo [1] physical engine for accurate robot-object interactions. Our system also includes easy-to-use ROS2 (Rob… view at source ↗
Figure 3
Figure 3. DISCOVERSE Real2Sim generation pipeline. We use 3DGS as a universal visual representation and integrate laser scanning, state-of-the-art generative models, and physically-based relighting to boost the geometry and appearance fidelity of the reconstructed radiance fields. B. Interactive Scene Nodes Real2Sim Generation As described in Sec. III, interactive scene nodes require a dual 3DGS-Mesh representation to support… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Visualizations of an AIRBOT Play robotic arm performing three different manipulation tasks in the simulation of D [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visualizations of an agent exploring a large-scale indoor scene in [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Visualizations of a quadrocopter and a wheeled loco-manipulator [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. AeroAct: Action-Centered World-Action Models for Language-Conditioned Quadrotor Flight

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A video-diffusion world-action model predicts smooth local trajectory chunks for language-conditioned quadrotor flight and is demonstrated in simulation and one real indoor flight.

  2. GS-Playground: A High-Throughput Photorealistic Simulator for Vision-Informed Robot Learning

    cs.RO 2026-04 unverdicted novelty 6.0 of 10

    GS-Playground delivers a high-throughput photorealistic simulator for vision-informed robot learning via parallel physics integrated with batch 3D Gaussian Splatting at 10^4 FPS and an automated Real2Sim workflow for ...

  3. From Seeing to Simulating: Generative High-Fidelity Simulation with Digital Cousins for Generalizable Robot Learning and Evaluation

    cs.RO 2026-04 unverdicted novelty 6.0 of 10

    Digital Cousins is a generative real-to-sim method that creates diverse high-fidelity simulation scenes from real panoramas to improve generalization in robot learning and evaluation.

  4. Genie Sim 3.0 : A High-Fidelity Comprehensive Simulation Platform for Humanoid Robot

    cs.RO 2026-01 unverdicted novelty 6.0 of 10

    Genie Sim 3.0 introduces an LLM-powered scene generator, the first LLM-based automated evaluation benchmark, and a large open synthetic dataset that demonstrates zero-shot sim-to-real transfer for robotic manipulation...

  5. Genie Sim 3.0 : A High-Fidelity Comprehensive Simulation Platform for Humanoid Robot

    cs.RO 2026-01 conditional novelty 6.0 of 10

    An open-source simulation platform with LLM-generated scenes, VLM-based automatic evaluation, 10,000+ hours of synthetic robot data, and evidence of sim-to-real transfer on four manipulation tasks.

  6. IGen: Scalable Data Generation for Robot Learning from Open-World Images

    cs.RO 2025-12 unverdicted novelty 6.0 of 10

    IGen generates realistic visuomotor training data including actions and temporally coherent visuals from unstructured open-world images via 3D reconstruction and VLM reasoning.

  7. SkillPlug: Unsupervised Skill Mining for Few-Shot Adaptation in Robotic Manipulation

    cs.RO 2026-07 conditional novelty 5.0 of 10

    Unsupervised skill mining with self-supervised compactness, alignment, and disentanglement losses yields a fixed skill library that improves multi-task and few-shot robotic manipulation when plugged into ACT and OpenVLA-OFT.

  8. QuadVerse: An Integrated Framework Aligning Visual-Physical Reality for Quadruped Simulation

    cs.RO 2026-06 unverdicted novelty 5.0 of 10

    QuadVerse integrates 3D Gaussian Splatting scene reconstruction, friction calibration via trajectory search, and a residual dynamics compensator to improve quadruped simulation fidelity and enable zero-shot policy transfer.

  9. REAP: Reinforcement-Learning End-to-End Autonomous Parking with Gaussian Splatting Simulator for Real2Sim2Real Transfer

    cs.RO 2026-05 unverdicted novelty 5.0 of 10

    REAP trains an end-to-end SAC policy with behavior cloning and collision penalties inside a 3DGS Real2Sim simulator and transfers it to physical vehicles, succeeding in narrow mechanical parking slots.

Reference graph

Works this paper leans on

46 extracted references · 21 canonical work pages · cited by 8 Pith papers

  1. [16]

    Robogsim: A real2sim2real robotic gaussian splatting simulator,

    X. Li, J. Li, Z. Zhang, R. Zhang, F. Jia, T. Wang, H. Fan, K.-K. Tseng, and R. Wang, “Robogsim: A real2sim2real robotic gaussian splatting simulator,” arXiv preprint arXiv:2411.11839 , 2024

    arXiv 2024

  2. [17]

    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. Shi et al., “Robo-gs: A physics consistent spatial-temporal model for robotic arm with hybrid representation,” arXiv preprint arXiv:2408.14873, 2024

    arXiv 2024

  3. [1]

    Mujoco: A physics engine for model-based control,

    E. Todorov, T. Erez, and Y . Tassa, “Mujoco: A physics engine for model-based control,” in 2012 IEEE/RSJ international conference on intelligent robots and systems . IEEE, 2012, pp. 5026–5033

    2012

  4. [2]

    Habitat 2.0: Training home assistants to rearrange their habitat,

    A. Szot, A. Clegg, E. Undersander, E. Wijmans, Y . Zhao, J. Turner, N. Maestre, M. Mukadam, D. S. Chaplot, O. Maksymets et al. , “Habitat 2.0: Training home assistants to rearrange their habitat,” Advances in neural information processing systems , vol. 34, pp. 251– 266, 2021

    work page 2021

  5. [3]

    Orbit: A unified simulation framework for interactive robot learning environments,

    M. Mittal, C. Yu, Q. Yu, J. Liu, N. Rudin, D. Hoeller, J. L. Yuan, R. Singh, Y . Guo, H. Mazhar et al. , “Orbit: A unified simulation framework for interactive robot learning environments,”IEEE Robotics and Automation Letters , vol. 8, no. 6, pp. 3740–3747, 2023

    work page 2023

  6. [4]

    Sapien: A simulated part-based interactive environment,

    F. Xiang, Y . Qin, K. Mo, Y . Xia, H. Zhu, F. Liu, M. Liu, H. Jiang, Y . Yuan, H. Wang et al. , “Sapien: A simulated part-based interactive environment,” in Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , 2020, pp. 11 097–11 107

    work page 2020

  7. [5]

    Three- dworld: A platform for interactive multi-modal physical simulation,

    C. Gan, J. Schwartz, S. Alter, D. Mrowca, M. Schrimpf, J. Traer, J. De Freitas, J. Kubilius, A. Bhandwaldar, N. Haber et al. , “Three- dworld: A platform for interactive multi-modal physical simulation,” arXiv preprint arXiv:2007.04954 , 2020

    arXiv 2007

  8. [6]

    Robotwin: Dual-arm robot benchmark with generative digital twins (early version),

    Y . Mu, T. Chen, S. Peng, Z. Chen, Z. Gao, Y . Zou, L. Lin, Z. Xie, and P. Luo, “Robotwin: Dual-arm robot benchmark with generative digital twins (early version),” arXiv preprint arXiv:2409.02920 , 2024

    arXiv 2024

Show all 46 references
  1. [7]

    Maniskill2: A unified benchmark for generalizable manipulation skills,

    J. Gu, F. Xiang, X. Li, Z. Ling, X. Liu, T. Mu, Y . Tang, S. Tao, X. Wei, Y . Yao et al. , “Maniskill2: A unified benchmark for generalizable manipulation skills,” arXiv preprint arXiv:2302.04659 , 2023

    2023 arXiv

  2. [8]

    Manipulathor: A framework for vi- sual object manipulation,

    K. Ehsani, W. Han, A. Herrasti, E. VanderBilt, L. Weihs, E. Kolve, A. Kembhavi, and R. Mottaghi, “Manipulathor: A framework for vi- sual object manipulation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2021, pp. 4497–4506

    2021

  3. [9]

    igibson 2.0: Object- centric simulation for robot learning of everyday household tasks,

    C. Li, F. Xia, R. Mart ´ın-Mart´ın, M. Lingelbach, S. Srivastava, B. Shen, K. Vainio, C. Gokmen, G. Dharan, T. Jain et al., “igibson 2.0: Object- centric simulation for robot learning of everyday household tasks,” arXiv preprint arXiv:2108.03272 , 2021

    2021 arXiv

  4. [10]

    Sim2real in robotics and automation: Applications and challenges,

    S. H ¨ofer, K. Bekris, A. Handa, J. C. Gamboa, M. Mozifian, F. Golemo, C. Atkeson, D. Fox, K. Goldberg, J. Leonard et al. , “Sim2real in robotics and automation: Applications and challenges,” IEEE trans- actions on automation science and engineering , vol. 18, no. 2, pp. 398–400, 2021

    2021

  5. [11]

    Navigating to objects in the real world,

    T. Gervet, S. Chintala, D. Batra, J. Malik, and D. S. Chaplot, “Navigating to objects in the real world,” Science Robotics , vol. 8, no. 79, p. eadf6991, 2023

    2023

  6. [12]

    Matterport3d: Learning from rgb-d data in indoor environments,

    A. Chang, A. Dai, T. Funkhouser, M. Halber, M. Niessner, M. Savva, S. Song, A. Zeng, and Y . Zhang, “Matterport3d: Learning from rgb-d data in indoor environments,” arXiv preprint arXiv:1709.06158 , 2017

    2017 arXiv

  7. [13]

    The replica dataset: A digital replica of indoor spaces,

    J. Straub, T. Whelan, L. Ma, Y . Chen, E. Wijmans, S. Green, J. J. Engel, R. Mur-Artal, C. Ren, S. Verma et al. , “The replica dataset: A digital replica of indoor spaces,” arXiv preprint arXiv:1906.05797 , 2019

    1906 arXiv

  8. [14]

    Habitat: A platform for embodied ai research,

    M. Savva, A. Kadian, O. Maksymets, Y . Zhao, E. Wijmans, B. Jain, J. Straub, J. Liu, V . Koltun, J. Malik et al. , “Habitat: A platform for embodied ai research,” in Proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 9339–9347

    2019

  9. [15]

    Interactive gibson benchmark: A benchmark for interactive navigation in cluttered environments,

    F. Xia, W. B. Shen, C. Li, P. Kasimbeg, M. E. Tchapmi, A. Toshev, R. Mart´ın-Mart´ın, and S. Savarese, “Interactive gibson benchmark: A benchmark for interactive navigation in cluttered environments,” IEEE Robotics and Automation Letters , vol. 5, no. 2, pp. 713–720, 2020

    2020

  10. [18]

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

    M. N. Qureshi, S. Garg, F. Yand ´un, D. Held, G. Kantor, and A. Silwal, “Splatsim: Zero-shot sim2real transfer of rgb manipulation policies using gaussian splatting,” ArXiv, vol. abs/2409.10161, 2024

    2024 arXiv

  11. [19]

    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.” ACM Trans. Graph. , vol. 42, no. 4, pp. 139–1, 2023

    2023

  12. [20]

    Learning fine-grained bimanual manipulation with low-cost hardware,

    T. Z. Zhao, V . Kumar, S. Levine, and C. Finn, “Learning fine-grained bimanual manipulation with low-cost hardware,” arXiv preprint arXiv:2304.13705, 2023

    2023 arXiv

  13. [21]

    Diffusion policy: Visuomotor policy learning via ac- tion diffusion,

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song, “Diffusion policy: Visuomotor policy learning via ac- tion diffusion,” The International Journal of Robotics Research , p. 02783649241273668, 2023

    2023

  14. [22]

    Pybullet, a python module for physics simulation for games, robotics and machine learning,

    E. Coumans and Y . Bai, “Pybullet, a python module for physics simulation for games, robotics and machine learning,” 2016. [Online]. Available: https://pybullet.org/wordpress/

    2016

  15. [23]

    Activesplat: High-fidelity scene reconstruction through active gaussian splatting,

    Y . Li, Z. Kuang, T. Li, G. Zhou, S. Zhang, and Z. Yan, “Activesplat: High-fidelity scene reconstruction through active gaussian splatting,” arXiv preprint arXiv:2410.21955 , 2024

    2024 arXiv

  16. [24]

    Clay: A controllable large-scale generative model for creating high-quality 3d assets,

    L. Zhang, Z. Wang, Q. Zhang, Q. Qiu, A. Pang, H. Jiang, W. Yang, L. Xu, and J. Yu, “Clay: A controllable large-scale generative model for creating high-quality 3d assets,” ACM Transactions on Graphics (TOG), vol. 43, no. 4, pp. 1–20, 2024

    2024

  17. [25]

    Robot operating system 2: Design, architecture, and uses in the wild,

    S. Macenski, T. Foote, B. Gerkey, C. Lalancette, and W. Woodall, “Robot operating system 2: Design, architecture, and uses in the wild,” Science robotics, vol. 7, no. 66, p. eabm6074, 2022

    2022

  18. [26]

    Pixel- wise view selection for unstructured multi-view stereo,

    J. L. Sch ¨onberger, E. Zheng, J.-M. Frahm, and M. Pollefeys, “Pixel- wise view selection for unstructured multi-view stereo,” in Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part III 14 . Springer, 2016, pp...

    2016

  19. [27]

    Structured 3d latents for scalable and versatile 3d generation,

    J. Xiang, Z. Lv, S. Xu, Y . Deng, R. Wang, B. Zhang, D. Chen, X. Tong, and J. Yang, “Structured 3d latents for scalable and versatile 3d generation,” arXiv preprint arXiv:2412.01506 , 2024

    2024 arXiv

  20. [28]

    Lgm: Large multi-view gaussian model for high-resolution 3d content creation,

    J. Tang, Z. Chen, X. Chen, T. Wang, G. Zeng, and Z. Liu, “Lgm: Large multi-view gaussian model for high-resolution 3d content creation,” in European Conference on Computer Vision . Springer, 2025, pp. 1–18

    2025

  21. [29]

    Shapenet: An information-rich 3d model repository,

    A. X. Chang, T. Funkhouser, L. Guibas, P. Hanrahan, Q. Huang, Z. Li, S. Savarese, M. Savva, S. Song, H. Su et al. , “Shapenet: An information-rich 3d model repository,” arXiv preprint arXiv:1512.03012, 2015

    2015 arXiv

  22. [30]

    Partnet: A large-scale benchmark for fine-grained and hierarchical part-level 3d object understanding,

    K. Mo, S. Zhu, A. X. Chang, L. Yi, S. Tripathi, L. J. Guibas, and H. Su, “Partnet: A large-scale benchmark for fine-grained and hierarchical part-level 3d object understanding,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 909– 918

    2019

  23. [31]

    Objaverse-xl: A universe of 10m+ 3d objects,

    M. Deitke, R. Liu, M. Wallingford, H. Ngo, O. Michel, A. Kusupati, A. Fan, C. Laforte, V . V oleti, S. Y . Gadre et al. , “Objaverse-xl: A universe of 10m+ 3d objects,” Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  24. [32]

    Approximate convex decompo- sition for 3d meshes with collision-aware concavity and tree search,

    X. Wei, M. Liu, Z. Ling, and H. Su, “Approximate convex decompo- sition for 3d meshes with collision-aware concavity and tree search,” ACM Transactions on Graphics (TOG) , vol. 41, no. 4, pp. 1–18, 2022

    2022

  25. [33]

    Tacchi: A pluggable and low computational cost elastomer deformation simulator for optical tactile sensors,

    Z. Chen, S. Zhang, S. Luo, F. Sun, and B. Fang, “Tacchi: A pluggable and low computational cost elastomer deformation simulator for optical tactile sensors,” IEEE Robotics and Automation Letters , vol. 8, no. 3, pp. 1239–1246, 2023

    2023

  26. [34]

    Letsgo: Large-scale garage modeling and rendering via lidar-assisted gaussian primitives,

    J. Cui, J. Cao, Y . Zhong, L. Wang, F. Zhao, P. Wang, Y . Chen, Z. He, L. Xu, Y . Shi et al. , “Letsgo: Large-scale garage modeling and rendering via lidar-assisted gaussian primitives,” arXiv preprint arXiv:2404.09748, 2024

    2024 arXiv

  27. [35]

    Urban radiance fields,

    K. Rematas, A. Liu, P. P. Srinivasan, J. T. Barron, A. Tagliasacchi, T. Funkhouser, and V . Ferrari, “Urban radiance fields,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 12 932–12 942

    2022

  28. [36]

    Street gaussians for modeling dynamic urban scenes,

    Y . Yan, H. Lin, C. Zhou, W. Wang, H. Sun, K. Zhan, X. Lang, X. Zhou, and S. Peng, “Street gaussians for modeling dynamic urban scenes,” arXiv preprint arXiv:2401.01339 , 2024

    2024 arXiv

  29. [37]

    Diffusionlight: Light probes for free by painting a chrome ball,

    P. Phongthawee, W. Chinchuthakun, N. Sinsunthithet, V . Jampani, A. Raj, P. Khungurn, and S. Suwajanakorn, “Diffusionlight: Light probes for free by painting a chrome ball,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 98–108

    2024

  30. [38]

    Method for registration of 3-d shapes,

    P. J. Besl and N. D. McKay, “Method for registration of 3-d shapes,” in Sensor fusion IV: control paradigms and data structures , vol. 1611. Spie, 1992, pp. 586–606

    1992

  31. [39]

    Kinectfusion: Real-time dense surface mapping and tracking,

    R. A. Newcombe, S. Izadi, O. Hilliges, D. Molyneaux, D. Kim, A. J. Davison, P. Kohi, J. Shotton, S. Hodges, and A. Fitzgibbon, “Kinectfusion: Real-time dense surface mapping and tracking,” in 2011 10th IEEE international symposium on mixed and augmented reality. Ieee, 2011, pp...

    2011

  32. [40]

    Surface simplification using quadric error metrics,

    M. Garland and P. S. Heckbert, “Surface simplification using quadric error metrics,” in Proceedings of the 24th annual conference on Computer graphics and interactive techniques , 1997, pp. 209–216

    1997

  33. [41]

    Learning visual parkour from generated images,

    A. Yu, G. Yang, R. Choi, Y . Ravan, J. Leonard, and P. Isola, “Learning visual parkour from generated images,” in 8th Annual Conference on Robot Learning , 2024

    2024

  34. [42]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” in Proceedings of the IEEE/CVF international conference on computer vision , 2023, pp. 3836–3847

    2023

  35. [43]

    Gpt-4v(ision) system card,

    “Gpt-4v(ision) system card,” 2023. [Online]. Available: https: //api.semanticscholar.org/CorpusID:263218031

    2023

  36. [44]

    Raft: Recurrent all-pairs field transforms for optical flow,

    Z. Teed and J. Deng, “Raft: Recurrent all-pairs field transforms for optical flow,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II

    2020

  37. [45]

    Springer, 2020, pp. 402–419

    2020

  38. [46]

    Two-frame motion estimation based on polynomial expansion,

    G. Farneb ¨ack, “Two-frame motion estimation based on polynomial expansion,” in Image Analysis: 13th Scandinavian Conference, SCIA 2003 Halmstad, Sweden, June 29–July 2, 2003 Proceedings 13 . Springer, 2003, pp. 363–370

    2003

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.