arxiv: 2605.02708 · v1 · submitted 2026-05-04 · 💻 cs.RO · cs.CV

Recognition: 3 theorem links

· Lean Theorem

Temporally Consistent Object 6D Pose Estimation for Robot Control

Kateryna Zorina , Vojtech Priban , Mederic Fourmy , Josef Sivic , Vladimir Petrik

Authors on Pith no claims yet

Pith reviewed 2026-05-08 18:04 UTC · model grok-4.3

classification 💻 cs.RO cs.CV

keywords 6D pose estimationfactor graphstemporal consistencyrobot controlobject trackingvision-based controloutlier rejectionmotion models

0 comments

The pith

A factor graph with motion models and uncertainty estimates enforces temporal consistency in single-view 6D object poses for stable robot control.

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

Single-view RGB pose estimators have reached sufficient precision for vision-based robot control but lack the temporal consistency required to keep feedback loops stable. This paper introduces a factor graph that incorporates object motion models and explicitly estimates the uncertainty of each pose measurement. These elements are combined in an online optimization that also performs outlier rejection and smoothing. The result is claimed to raise accuracy on standard benchmarks while enabling reliable closed-loop tracking when a camera on a torque-controlled manipulator follows a moving object.

Core claim

The central claim is that a factor graph approach which integrates object motion models and estimated measurement uncertainties, solved through online optimization with outlier rejection, produces temporally consistent 6D pose estimates that measurably outperform single-view methods on benchmarks and support stable feedback control in a real robot experiment.

What carries the argument

The factor graph estimator that adds motion-model factors and uncertainty-weighted measurement factors, solved online with outlier rejection and smoothing.

Load-bearing premise

Accurate object motion models can be supplied and integrated in real time without bias or excessive latency, and the estimated uncertainties correctly reflect measurement quality so that factors receive proper weights.

What would settle it

Running the robot control experiment and finding that pose estimates remain inconsistent enough to cause unstable tracking, or seeing no accuracy gain on benchmarks relative to the raw single-view estimator.

Figures

Figures reproduced from arXiv: 2605.02708 by Josef Sivic, Kateryna Zorina, Mederic Fourmy, Vladimir Petrik, Vojtech Priban.

**Figure 1.** Figure 1: Mustard bottle object pose estimates from images. The plot (bottom) shows the angular distance between the estimated pose and a fixed reference frame. The shown objects are static, and therefore the distance should be constant. The red dots show the per-frame estimates computed by an object pose estimator CosyPose [2]. Filtered predictions computed by our method are shown in green. The corresponding red an… view at source ↗

**Figure 2.** Figure 2: Overview. Our goal is to estimate the poses of objects in time with respect to the reference frame R as shown in figure a. To achieve this, we use measurements at a time step k of the camera pose T˜k C and the object pose T˜k,A CO , where A is the label of the object. Both objects and the robot are moving in time, as illustrated by purple arrows. Our approach maintains the probabilistic world representatio… view at source ↗

**Figure 3.** Figure 3: Measurement covariance model. Visualization of the translation covariance model for the object pose estimations. Consider two objects (red and purple) whose projection on the image plane (dotted line) is shown in red and purple, respectively. The size of the covariance ellipsoid depends on the size of the object in the image plane. The uncertainty is higher in the direction of ray that points towards the o… view at source ↗

**Figure 5.** Figure 5: Qualitative results on SynthHOPEDynamic. Comparison between our method and CosyPose [2] on SynthHOPEDynamic sequence shown in the first row. The center of the frame is occluded by a black rectangle, and some of the frames are artificially blurred in the input video. It can be seen that some of the objects are not detected by per-frame CosyPose shown in the second row (e.g., frames 2 and 3) or that some o… view at source ↗

**Figure 6.** Figure 6: Ablation study for the constant pose motion model (left) evaluated on three scenes of the static synthetic dataset and for the constant velocity motion model (right) evaluated on three scenes of the dynamic synthetic dataset. The precisionrecall trade-off is controlled by hyperparameters of our model. Recall oriented parameters are selected such that recall is maximal and precision is at least at CosyPose… view at source ↗

**Figure 7.** Figure 7: The robot control architecture used for the tracking experiment. First, an image I k is used with CosyPose to generate object pose estimates. These estimates are then fed into the proposed Refiner along with the camera pose T k W C whose timestamp corresponds to the time stamp of the input image I k used in CosyPose. This synchronization is achieved by buffering the poses TW C and subsequently selecting th… view at source ↗

**Figure 9.** Figure 9: Analysis of robot tracking experiment. The evolution of the object angular distance for the robot tracking experiment. If the object is not occluded, CosyPose and our method predicts the object pose accurately (first frame). However, when object is completely occluded the per-frame evaluation cannot evaluate the pose of the object (second frame). Finally, if the object is partially visible, CosyPose predi… view at source ↗

read the original abstract

Single-view RGB object pose estimators have reached a level of precision and efficiency that makes them good candidates for vision-based robot control. However, off-the-shelf methods lack temporal consistency and robustness that are mandatory for a stable feedback control. In this work, we develop a factor graph approach to enforce temporal consistency of the object pose estimates. In particular, the proposed approach: (i) incorporates object motion models, (ii) explicitly estimates the object pose measurement uncertainty, and (iii) integrates the above two components in an online optimization-based estimator. We demonstrate that with appropriate outlier rejection and smoothing using the proposed factor graph approach, we can significantly improve the results on standardized pose estimation benchmarks. We experimentally validate the stability of the proposed approach for a feedback-based robot control task in which the object is tracked by the camera attached to a torque controlled manipulator.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper wraps single-view pose networks in an online factor graph with motion models and uncertainty to reduce jitter for robot control, delivering usable but incremental gains.

read the letter

The core contribution here is taking existing RGB 6D pose estimators and running them through a factor-graph optimizer that adds object motion models plus per-measurement uncertainty. This produces smoother trajectories that hold up better in a closed-loop manipulation task on a torque-controlled arm with the camera mounted on the robot itself. The abstract and approach are straightforward engineering: outlier rejection plus smoothing inside the graph, evaluated on standard benchmarks plus one real control experiment. That last part is the part worth noticing, because most pose papers never close the loop on hardware. The integration looks clean enough on paper and the assumptions (usable motion models, decent uncertainty estimates) are stated plainly. No obvious internal contradictions or circular definitions show up in the description. The work is honest about building on off-the-shelf networks rather than claiming a new pose architecture. The soft spots are modest but real. The claimed improvements depend heavily on the quality of the motion model and the accuracy of the uncertainty outputs; if either is weak the graph can add bias instead of removing noise. Without seeing the full ablations it is hard to separate how much comes from the factor graph versus the outlier rejection step. Direct comparisons to simpler alternatives like a basic Kalman filter on the same pose stream would help clarify whether the extra machinery is necessary. The paper is aimed at robotics people who already use single-view pose estimators but need more stability for feedback control. A reader working on manipulation or visual servoing would get practical takeaways from the control results. It is not a foundational methods paper, but the combination is specific enough and the hardware validation is concrete enough that it merits referee time. I would send it to review rather than desk reject.

Referee Report

2 major / 3 minor

Summary. The paper proposes a factor-graph estimator to enforce temporal consistency in single-view RGB 6D object pose estimation for robot control. It integrates object motion models, explicitly estimates per-measurement pose uncertainties, performs online optimization, and applies outlier rejection and smoothing. The approach is claimed to yield significant gains on standard pose benchmarks and to provide stable feedback control when a camera mounted on a torque-controlled manipulator tracks a moving object.

Significance. If the reported gains hold under the stated assumptions, the work offers a practical, modular way to add dynamics and uncertainty awareness to existing single-frame pose estimators, which is relevant for closed-loop robotic tasks. The hardware validation on a torque-controlled arm is a strength, as is the explicit use of factor graphs to combine motion priors with measurement factors.

major comments (2)

§3 (Factor Graph Formulation): The central claim that the estimator is 'parameter-free' once motion models and uncertainties are incorporated is not supported by the description of how the motion-model factors are instantiated; the choice of process noise covariance and the linearization point for the relative-pose factors appear to require tuning that is not shown to be independent of the object or scene.
Table 2 (Benchmark Results): The reported ADD-S and AUC improvements are presented without per-sequence standard deviations or statistical significance tests; given that the method adds temporal smoothing, it is unclear whether the gains exceed what could be obtained by a simple low-pass filter on the same single-frame poses.

minor comments (3)

§4.1: The description of the uncertainty estimation network does not specify the loss used for training the variance head or whether it is trained jointly with the pose estimator.
Figure 3: The legend for the factor-graph diagram is missing; it is unclear which edges correspond to motion-model factors versus measurement factors.
The abstract states that 'suitable object motion models exist'; this assumption should be stated explicitly as a limitation in the conclusion, together with a brief discussion of failure cases when the model is misspecified.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive assessment of the work's relevance for closed-loop robotic tasks. We address each major comment below and describe the revisions we will incorporate.

read point-by-point responses

Referee: §3 (Factor Graph Formulation): The central claim that the estimator is 'parameter-free' once motion models and uncertainties are incorporated is not supported by the description of how the motion-model factors are instantiated; the choice of process noise covariance and the linearization point for the relative-pose factors appear to require tuning that is not shown to be independent of the object or scene.

Authors: We acknowledge that the phrasing in §3 regarding the estimator being 'parameter-free' is imprecise and not fully supported by the provided details on factor instantiation. In the revised manuscript we will remove the 'parameter-free' claim. We will instead explicitly document the fixed process-noise covariance values (chosen once for the constant-velocity motion model based on typical object speeds in manipulation scenarios) and the linearization strategy (using the previous posterior as the operating point). These choices are applied uniformly across all objects and scenes in our experiments without per-instance tuning; we will add a short paragraph with the exact numerical values and a brief sensitivity discussion to demonstrate generality. revision: yes
Referee: Table 2 (Benchmark Results): The reported ADD-S and AUC improvements are presented without per-sequence standard deviations or statistical significance tests; given that the method adds temporal smoothing, it is unclear whether the gains exceed what could be obtained by a simple low-pass filter on the same single-frame poses.

Authors: We agree that the current presentation of Table 2 lacks sufficient statistical detail. In the revision we will augment the table with per-sequence standard deviations for ADD-S and AUC. We will also add a new ablation subsection that applies a simple low-pass filter (with comparable cutoff) to the identical single-frame baseline poses and reports the resulting metrics side-by-side with our factor-graph results. This will allow direct comparison and show that the observed gains derive from the joint optimization of motion priors, uncertainty-aware measurements, and outlier rejection rather than smoothing alone. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents a factor-graph formulation that integrates standard object motion models and per-measurement uncertainty estimates into an online pose estimator. No equation or algorithmic step reduces by construction to a quantity fitted inside the same paper; the claimed improvements on benchmarks and control stability follow from the explicit incorporation of these external components rather than from any self-definitional identity or renamed fit. The central result therefore remains independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities.

pith-pipeline@v0.9.0 · 5454 in / 981 out tokens · 55037 ms · 2026-05-08T18:04:29.183666+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Cost.FunctionalEquation (J(x)=½(x+x⁻¹)−1) washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

σ(n_px) = a·exp(−b·n_px), where a and b are parameters fitted separately for the translation xy, the translation z (i.e., depth) and the rotation

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

50 extracted references · 6 canonical work pages

[1]

Bop challenge 2023 on detection, segmentation and pose estimation of seen and unseen rigid objects,

T. Hodan, M. Sundermeyer, Y . Labbe, V . N. Nguyen, G. Wang, E. Brach- mann, B. Drost, V . Lepetit, C. Rother, and J. Matas, “Bop challenge 2023 on detection, segmentation and pose estimation of seen and unseen rigid objects,”arXiv:2403.09799, 2024

work page arXiv 2023
[2]

Cosypose: Consistent multi-view multi-object 6d pose estimation,

Y . Labb ´e, J. Carpentier, M. Aubry, and J. Sivic, “Cosypose: Consistent multi-view multi-object 6d pose estimation,” inECCV, 2020

2020
[3]

Megapose: 6d pose estimation of novel objects via render&compare,

Y . Labb ´e, L. Manuelli, A. Mousavian, S. Tyree, S. Birchfield, J. Trem- blay, J. Carpentier, M. Aubry, D. Fox, and J. Sivic, “Megapose: 6d pose estimation of novel objects via render&compare,” inCoRL, 2022

2022
[4]

Srt3d: A sparse region-based 3d object tracking approach for the real world,

M. Stoiber, M. Pfanne, K. H. Strobl, R. Triebel, and A. Albu-Sch ¨affer, “Srt3d: A sparse region-based 3d object tracking approach for the real world,”IJCV, 2022

2022
[5]

Simtrack: A simulation-based framework for scalable real-time object pose detection and tracking,

K. Pauwels and D. Kragic, “Simtrack: A simulation-based framework for scalable real-time object pose detection and tracking,” inIROS, 2015

2015
[6]

A tutorial on graph-based slam,

G. Grisetti, R. K ¨ummerle, C. Stachniss, and W. Burgard, “A tutorial on graph-based slam,”IEEE ITSM, 2010

2010
[7]

Recent advances in 3d object and hand pose estimation,

V . Lepetit, “Recent advances in 3d object and hand pose estimation,” arXiv:2006.05927, 2020

work page arXiv 2006
[8]

Bop challenge 2020 on 6d object localization,

T. Hoda ˇn, M. Sundermeyer, B. Drost, Y . Labb ´e, E. Brachmann, F. Michel, C. Rother, and J. Matas, “Bop challenge 2020 on 6d object localization,” inECCV Workshops, 2020

2020
[9]

Cnos: A strong baseline for cad-based novel object segmentation,

V . N. Nguyen, T. Groueix, G. Ponimatkin, V . Lepetit, and T. Hodan, “Cnos: A strong baseline for cad-based novel object segmentation,” in ICCV, 2023

2023
[10]

Gigapose: Fast and robust novel object pose estimation via one correspondence,

V . N. Nguyen, T. Groueix, M. Salzmann, and V . Lepetit, “Gigapose: Fast and robust novel object pose estimation via one correspondence,” inCVPR, 2024

2024
[11]

Foundpose: Unseen object pose estimation with foundation features,

E. P. ¨Ornek, Y . Labb´e, B. Tekin, L. Ma, C. Keskin, C. Forster, and T. Hodan, “Foundpose: Unseen object pose estimation with foundation features,”arXiv:2311.18809, 2023

work page arXiv 2023
[12]

FoundationPose: Unified 6D pose estimation and tracking of novel objects

B. Wen, W. Yang, J. Kautz, and S. Birchfield, “Foundationpose: Unified 6d pose estimation and tracking of novel objects,”arXiv:2312.08344, 2023

work page arXiv 2023
[13]

Deepim: Deep iterative matching for 6d pose estimation,

Y . Li, G. Wang, X. Ji, Y . Xiang, and D. Fox, “Deepim: Deep iterative matching for 6d pose estimation,” inECCV, 2018

2018
[14]

se (3)-tracknet: Data- driven 6d pose tracking by calibrating image residuals in synthetic domains,

B. Wen, C. Mitash, B. Ren, and K. E. Bekris, “se (3)-tracknet: Data- driven 6d pose tracking by calibrating image residuals in synthetic domains,” inIROS, 2020

2020
[15]

Poserbpf: A rao–blackwellized particle filter for 6-d object pose track- ing,

X. Deng, A. Mousavian, Y . Xiang, F. Xia, T. Bretl, and D. Fox, “Poserbpf: A rao–blackwellized particle filter for 6-d object pose track- ing,”TRO, 2021

2021
[16]

What uncertainties do we need in bayesian deep learning for computer vision?

A. Kendall and Y . Gal, “What uncertainties do we need in bayesian deep learning for computer vision?”Advances in neural information processing systems, 2017

2017
[17]

Survey on visual servoing for manipulation,

D. Kragic, H. I. Christensenet al., “Survey on visual servoing for manipulation,” 2002

2002
[18]

Quadricslam: Dual quadrics from object detections as landmarks in object-oriented slam,

L. Nicholson, M. Milford, and N. S ¨underhauf, “Quadricslam: Dual quadrics from object detections as landmarks in object-oriented slam,” RAL, 2018

2018
[19]

Cubeslam: Monocular 3-d object slam,

S. Yang and S. Scherer, “Cubeslam: Monocular 3-d object slam,”TRO, 2019

2019
[20]

Odam: Object detection, association, and mapping using posed rgb video,

K. Li, D. DeTone, Y . F. S. Chen, M. V o, I. Reid, H. Rezatofighi, C. Sweeney, J. Straub, and R. Newcombe, “Odam: Object detection, association, and mapping using posed rgb video,” inICCV, 2021

2021
[21]

Simultaneous localisation and mapping with quadric surfaces,

T. Laidlow and A. J. Davison, “Simultaneous localisation and mapping with quadric surfaces,” in3DV, 2022

2022
[22]

Fusion++: V olumetric object-level slam,

J. McCormac, R. Clark, M. Bloesch, A. Davison, and S. Leutenegger, “Fusion++: V olumetric object-level slam,” in3DV, 2018

2018
[23]

Morefusion: Multi-object reasoning for 6d pose estimation from volumetric fusion,

K. Wada, E. Sucar, S. James, D. Lenton, and A. J. Davison, “Morefusion: Multi-object reasoning for 6d pose estimation from volumetric fusion,” inCVPR, 2020

2020
[24]

Nodeslam: Neural object descrip- tors for multi-view shape reconstruction,

E. Sucar, K. Wada, and A. Davison, “Nodeslam: Neural object descrip- tors for multi-view shape reconstruction,” in3DV, 2020

2020
[25]

Simstack: A generative shape and instance model for unordered object stacks,

Z. Landgraf, R. Scona, T. Laidlow, S. James, S. Leutenegger, and A. J. Davison, “Simstack: A generative shape and instance model for unordered object stacks,” inICCV, 2021

2021
[26]

Slam++: Simultaneous localisation and mapping at the level of objects,

R. F. Salas-Moreno, R. A. Newcombe, H. Strasdat, P. H. Kelly, and A. J. Davison, “Slam++: Simultaneous localisation and mapping at the level of objects,” inCVPR, 2013

2013
[27]

A multi- hypothesis approach to pose ambiguity in object-based slam,

J. Fu, Q. Huang, K. Doherty, Y . Wang, and J. J. Leonard, “A multi- hypothesis approach to pose ambiguity in object-based slam,” inIROS, 2021

2021
[28]

Symmetry and uncertainty-aware object slam for 6dof object pose estimation,

N. Merrill, Y . Guo, X. Zuo, X. Huang, S. Leutenegger, X. Peng, L. Ren, and G. Huang, “Symmetry and uncertainty-aware object slam for 6dof object pose estimation,” inCVPR, 2022

2022
[29]

Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,

M. A. Fischler and R. C. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,”Communications of the ACM, 1981

1981
[30]

Bundle adjustment—a modern synthesis,

B. Triggs, P. F. McLauchlan, R. I. Hartley, and A. W. Fitzgibbon, “Bundle adjustment—a modern synthesis,” inInternational Workshop on Vision Algorithms, 2000

2000
[31]

Co-fusion: Real-time segmentation, tracking and fusion of multiple objects,

M. R ¨unz and L. Agapito, “Co-fusion: Real-time segmentation, tracking and fusion of multiple objects,” inICRA, 2017

2017
[32]

Maskfusion: Real-time recogni- tion, tracking and reconstruction of multiple moving objects,

M. Runz, M. Buffier, and L. Agapito, “Maskfusion: Real-time recogni- tion, tracking and reconstruction of multiple moving objects,” inISMAR, 2018

2018
[33]

Mid-fusion: Octree-based object-level multi-instance dynamic slam,

B. Xu, W. Li, D. Tzoumanikas, M. Bloesch, A. Davison, and S. Leutenegger, “Mid-fusion: Octree-based object-level multi-instance dynamic slam,” inICRA, 2019

2019
[34]

Moltr: Multiple object localization, tracking and reconstruction from monocular rgb videos,

K. Li, H. Rezatofighi, and I. Reid, “Moltr: Multiple object localization, tracking and reconstruction from monocular rgb videos,”RAL, 2021

2021
[35]

Learning to complete object shapes for object-level mapping in dynamic scenes,

B. Xu, A. J. Davison, and S. Leutenegger, “Learning to complete object shapes for object-level mapping in dynamic scenes,” inIROS, 2022

2022
[36]

Dynamic slam: The need for speed,

M. Henein, J. Zhang, R. Mahony, and V . Ila, “Dynamic slam: The need for speed,” inICRA, 2020

2020
[37]

Depth-based object tracking using a robust gaussian filter,

J. Issac, M. W ¨uthrich, C. G. Cifuentes, J. Bohg, S. Trimpe, and S. Schaal, “Depth-based object tracking using a robust gaussian filter,” inICRA, 2016

2016
[38]

Factor graphs for robot perception,

F. Dellaert, M. Kaesset al., “Factor graphs for robot perception,” F oundations and Trends® in Robotics, 2017

2017
[39]

The OpenCV Library,

G. Bradski, “The OpenCV Library,”Dr . Dobb’s Journal of Software Tools, 2000

2000
[40]

A micro lie theory f or state es timation in robotics,

J. Sol `a, J. Deray, and D. Atchuthan, “A micro lie theory for state estimation in robotics,”arXiv:1812.01537, 2021

work page arXiv 2021
[41]

Mask r-cnn,

K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” inICCV, 2017

2017
[42]

6-dof pose estimation of household objects for robotic manipulation: An accessible dataset and benchmark,

S. Tyree, J. Tremblay, T. To, J. Cheng, T. Mosier, J. Smith, and S. Birchfield, “6-dof pose estimation of household objects for robotic manipulation: An accessible dataset and benchmark,” inIROS, 2022

2022
[43]

The ycb object and model set: Towards common benchmarks for manipulation research,

B. Calli, A. Singh, A. Walsman, S. Srinivasa, P. Abbeel, and A. M. Dollar, “The ycb object and model set: Towards common benchmarks for manipulation research,” inICAR, 2015

2015
[44]

PoseCNN: A Convolutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes

Y . Xiang, T. Schmidt, V . Narayanan, and D. Fox, “Posecnn: A convolu- tional neural network for 6d object pose estimation in cluttered scenes,” arXiv:1711.00199, 2017. [45]Blender - a 3D modelling and rendering package, Blender Foundation,

work page Pith review arXiv 2017
[45]

Available: http://www.blender.org

[Online]. Available: http://www.blender.org
[46]

Orientation in cartesian space dynamic movement primitives,

A. Ude, B. Nemec, T. Petri ´c, and J. Morimoto, “Orientation in cartesian space dynamic movement primitives,” inICRA, 2014

2014
[47]

Bop: Benchmark for 6d object pose estimation,

T. Hodan, F. Michel, E. Brachmann, W. Kehl, A. GlentBuch, D. Kraft, B. Drost, J. Vidal, S. Ihrke, X. Zabuliset al., “Bop: Benchmark for 6d object pose estimation,” inECCV, 2018

2018
[48]

Rnnpose: Recurrent 6-dof object pose refinement with robust correspondence field estimation and pose optimization,

Y . Xu, K.-Y . Lin, G. Zhang, X. Wang, and H. Li, “Rnnpose: Recurrent 6-dof object pose refinement with robust correspondence field estimation and pose optimization,” inCVPR, 2022

2022
[49]

So- pose: Exploiting self-occlusion for direct 6d pose estimation,

Y . Di, F. Manhardt, G. Wang, X. Ji, N. Navab, and F. Tombari, “So- pose: Exploiting self-occlusion for direct 6d pose estimation,” inICCV, 2021

2021
[50]

A unified approach for motion and force control of robot manipulators: The operational space formulation,

O. Khatib, “A unified approach for motion and force control of robot manipulators: The operational space formulation,”Journal on Robotics and Automation, 1987

1987