Pith. sign in

REVIEW 3 major objections 7 minor 185 references

Learned policy replaces hand-tuning in distributed robot optimization

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 · glm-5.2

2026-07-10 02:05 UTC pith:MUCANFRK

load-bearing objection Solid L2O approach to distributed Lie-group optimization; train-test dynamics mismatch is a real but non-fatal gap the 3 major comments →

arxiv 2607.08735 v1 pith:MUCANFRK submitted 2026-07-09 cs.RO

Learning Adaptive Solvers for Distributed Factor Graph Optimization on Matrix Lie Groups

Jaeho Shin , Maani Ghaffari , Yulun Tian This is my paper
classification cs.RO
keywords distributedoptimizationgraphgroupsmatrixacrossadaptivecommunication
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.

The paper claims that the parameters governing distributed geometric optimization over matrix Lie groups can be replaced by a learned feedback policy. Instead of manually setting mass, damping, and step size for a Riemannian optimizer, the authors unfold the optimizer's iterations into a differentiable computation graph and train a graph neural network to predict these parameters from local optimization context. The network sees local velocities, gradients, residuals, and communication delay statistics, and outputs per-robot, per-iteration parameters. Trained self-supervised by minimizing the optimization cost along 50 unrolled iterations, the resulting solver generalizes to graphs of 10,000+ nodes, 500+ iterations, and communication regimes with random delays and packet loss. On 21 of 26 pose graph benchmarks and all three projective alignment benchmarks, it achieves lower objective values than hand-tuned distributed baselines.

Core claim

The central mechanism is the replacement of fixed solver hyperparameters with a shared, locally-computed feedback policy. Each robot constructs a two-hop augmented graph from its own variables and the most recently received neighbor states, attaches node-level features (body velocity, gradient, communication staleness), edge-level features (residuals, precision weights), and graph-level features (intra/inter-robot residual norms, update magnitudes), and feeds these through a GPS graph transformer with a parameter prediction head. The predicted mass, damping, and step size are then consumed by the underlying Riemannian optimizer update. The key empirical finding is that this policy, trained只在

What carries the argument

The load-bearing machinery is the deep unfolding of the CORD Riemannian optimizer: each iteration's parameter selection is made differentiable, gradients of the final optimization cost flow back through 50 iterations of solver steps, and a graph neural network learns to map local optimization state to solver parameters. Implicit differentiation through a preconditioned conjugate gradient solver avoids memory blowup from backpropagating through linear solves.

Load-bearing premise

The central assumption is that a policy trained on small graphs (~500 nodes) and short horizons (50 iterations) will generalize to graphs with 10,000+ nodes and 500+ iterations. The paper acknowledges this limitation but the empirical results on the Rim dataset provide the main evidence that it holds.

What would settle it

If the learned policy fails to outperform well-tuned fixed parameters on problem instances whose topology, noise level, or communication regime differs substantially from training data, the core claim of robust generalization would be undermined. The CampusRoad3 synchronous result, where the policy enters premature refinement due to small local gradients despite large global error, is a concrete instance of this failure mode.

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

If this is right

  • If a learned policy can replace hand-tuned parameters for distributed Lie group optimization, the same approach could extend to other distributed iterative solvers where parameter sensitivity is the main deployment bottleneck.
  • The finding that parameters trained on 500-node, 50-iteration graphs transfer to 10,000-node, 500-iteration settings suggests that optimization dynamics have transferable structure that a policy can capture without problem-specific retraining.
  • The phase-specific parameter schedules discovered by the policy (conservative early, aggressive late, communication-aware) provide empirical priors that could inform the design of non-learned adaptive solvers.
  • Generalization beyond SE(3) to SL(4) projective alignment with a separate trained model suggests the framework is group-agnostic in principle, though whether a single policy could span multiple groups remains open.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper trains separate models for SE(3) and SL(4), leaving open whether a unified policy across Lie groups is feasible or whether group-specific dynamics require separate learned components.
  • The conservatism on CampusRoad3 (where small local gradients cause premature refinement) suggests that purely local features may be fundamentally insufficient for detecting global optimality, and that some form of global or dual information may be necessary for robust convergence detection.
  • The energy dissipation argument for the simplified training dynamics (Appendix D) suggests that the learned parameters preserve convergence properties of the underlying ODE, but this is shown only for the simplified system used during training, not for the full dynamics used at inference.

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

3 major / 7 minor

Summary. The paper presents DeepCORD, a learning-augmented framework for distributed factor graph optimization over general matrix Lie groups. The method unfolds the CORD distributed Riemannian optimizer into differentiable iterations and trains a GNN-based feedback policy to predict adaptive solver parameters (mass, damping, step size) from local optimization context. Training is self-supervised via deep unfolding, minimizing unrolled objectives without requiring optimal solutions. The method is evaluated on SE(3) pose graph optimization (13 benchmark sequences under synchronous and asynchronous communication) and SL(4) projective submap alignment (3 TUM RGB-D sequences), demonstrating lower objective values than distributed baselines on most benchmarks. The approach is principled in preserving the geometric structure of the underlying optimizer while replacing manual parameter tuning with a learned policy.

Significance. The problem addressed—parameter sensitivity in distributed geometric optimization—is practically important for multi-robot SLAM and 3D reconstruction systems. The contribution of extending learning-to-optimize to distributed optimization over general matrix Lie groups, including under asynchronous communication, is novel and fills a genuine gap in the literature. The self-supervised training formulation (Eq. 7) that avoids dependence on optimal solutions is a strength, as is the demonstration of generalization to graphs larger than those in training (Rim: 10,195 nodes vs. training at ~500 nodes). The extension beyond SE(3) to SL(4) projective alignment, motivated by recent feed-forward SLAM systems, is timely. The extensive experimental evaluation across 26 PGO instances and 3 SL(4) datasets, with both synchronous and asynchronous protocols, provides substantial empirical support for the central claim.

major comments (3)
  1. Train-test dynamics mismatch (§4, Eq. 4 vs. Eq. 8/Appendix D). The policy is trained by unrolling a simplified ODE (Eq. 8) that omits the co-adjoint term ad*_ξ(Mξ) and the time-varying mass term Ṁξ, but at test time the full CORD dynamics (Eq. 4) are used. The co-adjoint term captures Lie-group non-commutativity and is potentially non-negligible for SE(3) (rotation-translation coupling) and especially SL(4) (15-dimensional algebra). Appendix D shows the simplified dynamics preserves energy dissipation qualitatively, but this does not establish that parameters optimal for the simplified dynamics remain near-optimal under the full dynamics. The paper should provide at least one of the following: (a) an ablation comparing test-time performance when the policy is trained on full vs. simplified dynamics, (b) empirical evidence that state distributions (ξ, gradient, residuals) encountered at测试
  2. SL(4) results show a large gap to centralized reference (Table 2). On TUM1, DeepCORD achieves cost 2.304 vs. reference 0.876—a gap of ~2.6×. While the paper attributes this to 'severe initial perturbations' (0.2 rad rotation, 0.1 m translation), it is unclear whether the dynamics mismatch described above contributes to this gap, or whether the gap reflects a fundamental limitation of the distributed approach under such perturbations. A sensitivity analysis varying the perturbation magnitude on SL(4) (analogous to Appendix A.3 for SE(3)) would help disentangle these factors and clarify whether the learned policy is limited by the training simplification or by problem difficulty.
  3. Generalization horizon (§4, §7). The policy is trained on K=50 unrolled iterations but evaluated over 500 iterations (Fig. 5). While the performance profiles show stable convergence, the paper does not report whether the predicted parameters remain well-behaved beyond 100 iterations or whether the policy enters degenerate regimes (e.g., vanishing step sizes) over very long horizons. Reporting the distribution of predicted parameters over the full 500-iteration horizon (extending Fig. 6) would strengthen the claim of stable long-horizon generalization.
minor comments (7)
  1. §4, Eq. (6): The policy outputs (m_k^i, d_k^i, Δt_k^i) are robot-level scalars, but the CORD update in Eq. (4) uses block-diagonal M_i = mH_i and D_i = dH_i where H_i varies per robot. It would help to clarify whether the policy predicts a single scalar per parameter per robot, or per-node values that are aggregated.
  2. Table 1: The 'Ref.' column is described as obtained from SE-Sync, but for some datasets (e.g., S3E sequences) the reference cost is much lower than any distributed method. Clarifying whether SE-Sync was run to global optimality or to a tolerance would help interpret the gaps.
  3. Appendix B.1: The communication feature c_u^k for asynchronous communication is defined as qΔt_{k-q}^j, but the units and interpretation are unclear. Is this the product of delay (in iterations) and the delayed step size? A brief clarification would help reproducibility.
  4. §5.2: The SL(4) evaluation uses only 3 datasets with relatively small graphs (89–249 nodes). Given that the SE(3) evaluation includes graphs up to 10,195 nodes, the SL(4) evaluation would benefit from at least one larger-scale example to support the generalization claim.
  5. Figure 4: The architecture diagram is dense and some labels are difficult to read. Consider simplifying or enlarging key components for clarity.
  6. Appendix C.1: The downsampling procedure for real-trajectory graphs is described in detail, but the rationale for the specific thresholds (5 m for loop closures, 1.4 m for grid/helix, etc.) is not provided. A brief justification would improve reproducibility.
  7. §7 (Limitations): The discussion of overly conservative behavior on CampusRoad3 is useful but brief. Quantifying the 'premature refinement regime' (e.g., showing the predicted parameter trajectories for this specific case) would make the limitation more concrete and actionable.

Circularity Check

0 steps flagged

No significant circularity; self-citation to CORD is the base algorithm being extended, not a theorem invoked to force the conclusion.

full rationale

The paper's derivation chain is: (1) take CORD [45] dynamics (Eq. 3–5), co-authored by present authors; (2) unfold iterations into a differentiable computation graph; (3) learn a policy π_θ that predicts solver parameters (m, d, Δt) via self-supervised loss minimizing the actual cost function along unrolled iterates (Eq. 7); (4) evaluate on external benchmarks not seen during training. The self-citation to CORD [45] is load-bearing in the sense that DeepCORD unfolds CORD's iterations, but this is standard practice in learning-to-optimize: you extend an existing algorithm by learning its hyperparameters. CORD itself is independently validated against external baselines in [45], and the present paper's central contribution—learning adaptive parameters—is independent of CORD's theoretical properties. The training loss (Eq. 7) minimizes the actual optimization objective evaluated along unrolled iterates, not a proxy or fitted quantity, so there is no fitted-input-called-prediction circularity. The simplified ODE used during training (Eq. 8 / Appendix D) is a modeling approximation transparently disclosed and justified by an energy dissipation argument, not an ansatz smuggled through citation. The skeptic's concern about train-test dynamics mismatch (simplified vs. full CORD dynamics) is a correctness/generalization risk, not a circularity issue. No step in the derivation reduces to its own inputs by construction.

Axiom & Free-Parameter Ledger

10 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities or mathematical objects. It uses established components: matrix Lie groups, graph neural networks (GPS), and the CORD optimizer. The free parameters are standard hyperparameters for neural network training and regularization.

free parameters (10)
  • Monotonicity regularization weight λ_b = 1.0
    Manually set regularization weight for penalizing cost increase between consecutive iterations (Tab. 6).
  • Damping regularization weight λ_d = 5.0
    Manually set regularization weight for encouraging decaying damping (Tab. 6).
  • Damping decay rate β = 0.5
    Constant in the damping regularization term (Tab. 6).
  • Damping decay offset τ = 5.0
    Constant in the damping regularization term (Tab. 6).
  • Number of unrolled iterations K = 50
    Training horizon, chosen due to memory limitations (Sec. 4).
  • Learning rate = 5e-5
    Adam optimizer learning rate (Sec. C.2).
  • Batch size = 16
    Training batch size (Sec. C.2).
  • GPS hidden dimension h = 128
    Latent space dimension for graph neural network (Sec. B.2).
  • Number of GPS layers = 4
    Depth of the graph transformer backbone (Sec. B.3).
  • Minimum parameter values = 1e-2
    Lower bound for mass and damping via softplus offset (Sec. B.4).
axioms (4)
  • domain assumption CORD dynamics (Eq. 3) converge to local minimizers with suitable parameters
    The paper builds on the theoretical foundation of CORD [45], assuming its convergence properties hold. This is invoked in Sec. 4 when describing the background.
  • domain assumption Simplified ODE (Eq. 8) preserves energy dissipation
    The paper proves in Appendix D that omitting co-adjoint and time-varying mass terms retains energy dissipation. This is used for training stability (Sec. 4).
  • ad hoc to paper Local graph features capture sufficient optimization context
    The policy predicts parameters from local augmented graph features. The paper acknowledges in Sec. 7 that local features may not fully capture proximity to global optimality.
  • ad hoc to paper Training on 50 iterations generalizes to 500+ iterations
    The training uses K=50 unrolled iterations but evaluation runs for 100-500 iterations. Performance profiles in Fig. 5 provide empirical support.

pith-pipeline@v1.1.0-glm · 21642 in / 2537 out tokens · 363584 ms · 2026-07-10T02:05:27.209942+00:00 · methodology

0 comments
read the original abstract

Modern robotic perception increasingly involves large-scale geometric optimization problems distributed across multiple robots or sessions. However, existing distributed solvers often depend on brittle hand tuning and primarily target rigid body pose graphs. To address this, we present DeepCORD, a learning-augmented framework for distributed factor graph optimization on general matrix Lie groups. By unfolding a parallel and accelerated Riemannian optimizer into differentiable iterations, DeepCORD learns a self-supervised feedback policy that dynamically adapts solver parameters according to the optimization phase and communication status. The resulting method enables adaptive distributed optimization over matrix Lie groups under both synchronous and asynchronous communication regimes. Extensive experiments on real-world $\mathrm{SE}$(3) pose graph optimization and $\mathrm{SL}$(4) projective submap alignment show that our method achieves lower objective values than existing distributed baselines on most benchmarks across realistic operating scenarios.

Figures

Figures reproduced from arXiv: 2607.08735 by Jaeho Shin, Maani Ghaffari, Yulun Tian.

Figure 1
Figure 1. Figure 1: DeepCORD aligns feed-forward SLAM submaps from four robots through distributed SL(4) optimization, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (Left) Node and edge partitioning in distributed opti￾mization. Colors indicate differ￾ent robots. (Right) The augmented graph for the red robot. In the distributed setting, the nodes are partitioned across a set of robots A = {1, . . . , N}. Robot i ∈ A owns a subset of variables Vi ⊆ V and maintains the corresponding local states Xi := {Xℓ}ℓ∈Vi . Each robot only optimizes the variables in its own subgrap… view at source ↗
Figure 3
Figure 3. Figure 3: Distributed PGO solved by the proposed method. Each color denotes a different robot. Pose Graph Optimization (PGO) over SE(3). State-of-the-art multi-robot SLAM systems use PGO to correct accumulated odometry drift and improve global consistency of trajectory estimation [24, 25, 53]; see [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Proposed architecture of the adaptive local feedback policy [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Evolution of the adaptive parameters (m, d, ∆t) predicted by DeepCORD over 100 iterations, illustrated across optimization phases and maximum delays. Analysis of Adaptive Parame￾ters. To provide additional insights, we analyze the DeepCORD parameter predictions on the largest Rim dataset under varied maximum delays ranging from 2 to 10 iterations. The 100 it￾erations are equally divided into early, middle,… view at source ↗
Figure 5
Figure 5. Figure 5: Performance profiles of PGO over 500 iterations (∆ = 0 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Fused maps from DeepCORD on CampusRoad1 (left), Dormitory1 (middle), and Library1 (right) after PGO. Each color denotes a different robot. This section presents the reconstructed point cloud maps from three representative sequences of the S3E dataset [1] as additional qualitative results for DeepCORD. As shown in [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Fused maps from DeepCORD on TUM1 (left), TUM2 (middle), and TUM3 (right) after projective submap [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Optimality gap comparison on the Rim dataset under varying initial perturbations. Dots represent the mean over 10 independent runs, and the shaded areas indicate the standard deviation. A.3 Robustness to Initialization Perturbations This section further evaluates the robustness of DeepCORD under varying initialization quality. Starting from the chordal initialization on the Rim PGO dataset, we perturb the … 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

185 extracted references · 185 canonical work pages · 14 internal anchors

  1. [1]

    Shin, Sungho , month = may, year =. Graph-. 2023. doi:10.23919/ACC55779.2023.10155947 , language =

    work page doi:10.23919/acc55779.2023.10155947 2023

  2. [2]

    Overlapping

    Shin, Sungho and Anitescu, Mihai and Zavala, Victor , pages =. Overlapping. 2020 59th. doi:10.1109/CDC42340.2020.9304139 , abstract =

    work page doi:10.1109/cdc42340.2020.9304139 2020

  3. [3]

    Na, Sen and Shin, Sungho and Anitescu, Mihai and Zavala, Victor , journal=. On the. 2022 , month=mar, volume=

    work page 2022

  4. [4]

    Exponential Decay of Sensitivity in Graph-Structured Nonlinear Programs

    Shin, Sungho and Anitescu, Mihai and Zavala, Victor , month = dec, year =. Exponential. doi:10.48550/arXiv.2101.03067 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2101.03067

  5. [5]

    IEEE Trans.\ on Control of Network Systems , author =

    Decentralized. IEEE Trans.\ on Control of Network Systems , author =. 2020 , keywords =. doi:10.1109/TCNS.2020.2967805 , abstract =

    work page doi:10.1109/tcns.2020.2967805 2020

  6. [6]

    IEEE Int

    McGann, Daniel and Lassak, Kyle and Kaess, Michael , file =. IEEE Int. Conf. on Robotics and Automation , title =

    work page

  7. [7]

    McGann and M

    D. McGann and M. Kaess , fullauthor =. Proc. Robotics: Science and Systems (RSS) , address =

    work page

  8. [11]

    and Dellaert, Frank , month = sep, year =

    Choudhary, Siddharth and Carlone, Luca and Christensen, Henrik I. and Dellaert, Frank , month = sep, year =. Exactly sparse memory efficient. 2015. doi:10.1109/IROS.2015.7353543 , abstract =

    work page doi:10.1109/iros.2015.7353543 2015

  9. [12]

    2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Cross-agent relocalization for decentralized collaborative SLAM , author=. 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2023 , organization=

    work page 2023

  10. [13]

    Distributed Pose Graph Optimization via Contractive Belief Sharing , year=

    Liu, Xiangyu and Chli, Margarita , journal=. Distributed Pose Graph Optimization via Contractive Belief Sharing , year=

    work page

  11. [14]

    GeoD: Consensus-based Geodesic Distributed Pose Graph Optimization

    Cristofalo, Eric and Montijano, Eduardo and Schwager, Mac , month = oct, year =. doi:10.48550/arXiv.2010.00156 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2010.00156 2010

  12. [15]

    Initialization techniques for

    Carlone, Luca and Tron, Roberto and Daniilidis, Kostas and Dellaert, Frank , pages =. Initialization techniques for. 2015. doi:10.1109/ICRA.2015.7139836 , abstract =

    work page doi:10.1109/icra.2015.7139836 2015

  13. [16]

    Generalized

    Fan, Taosha and Murphey, Todd , editor =. Generalized. Robotics. 2022 , doi =

    work page 2022

  14. [17]

    Decentralization and

    Fan, Taosha and Ortiz, Joseph and Hsiao, Ming and Monge, Maurizio and Dong, Jing and Murphey, Todd and Mukadam, Mustafa , month = jul, year =. Decentralization and. Robotics:. doi:10.15607/RSS.2023.XIX.111 , abstract =

    work page doi:10.15607/rss.2023.xix.111 2023

  15. [18]

    Efficient

    Konolige, K and Grisetti, G and Kümmerle, Rainer and Burgard, W and Limketkai, B and Vincent, R , pages =. Efficient. 2010. doi:10.1109/IROS.2010.5649043 , abstract =

    work page doi:10.1109/iros.2010.5649043 2010

  16. [19]

    A fast and accurate approximation for planar pose graph optimization , volume =. The Int. Journal of Robotics Research , author =. 2014 , pages =. doi:10.1177/0278364914523689 , abstract =

    work page doi:10.1177/0278364914523689 2014

  17. [20]

    Wei, Ermin and Ozdaglar, Asuman , month = dec, year =. On the. 2013. doi:10.1109/GlobalSIP.2013.6736937 , language =

    work page doi:10.1109/globalsip.2013.6736937 2013

  18. [21]

    Distributed

    Wei, Ermin and Ozdaglar, Asuman , month = dec, year =. Distributed. 2012. doi:10.1109/CDC.2012.6425904 , language =

    work page doi:10.1109/cdc.2012.6425904 2012

  19. [22]

    IEEE Trans

    Present and. IEEE Trans. on Robotics , author =. 2024 , keywords =. doi:10.1109/TRO.2023.3323938 , abstract =

    work page doi:10.1109/tro.2023.3323938 2024

  20. [23]

    Halsted, Trevor and Shorinwa, Ola and Yu, Javier and Schwager, Mac , month = mar, year =. A. doi:10.48550/arXiv.2103.12840 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2103.12840

  21. [24]

    and Kumar, V

    Karypis, G. and Kumar, V. , month = nov, year =. Multilevel. doi:10.1109/SC.1998.10018 , abstract =

    work page doi:10.1109/sc.1998.10018 1998

  22. [25]

    Iterative

    Saad, Yousef , year =. Iterative

    work page

  23. [26]

    IEEE Robotics and Automation Letters , author =

    Distributed. IEEE Robotics and Automation Letters , author =. 2024 , keywords =. doi:10.1109/LRA.2024.3382531 , abstract =

    work page doi:10.1109/lra.2024.3382531 2024

  24. [27]

    doi:10.48550/arXiv.2406.17249 , abstract =

    Liu, Xu and Lei, Jiuzhou and Prabhu, Ankit and Tao, Yuezhan and Spasojevic, Igor and Chaudhari, Pratik and Atanasov, Nikolay and Kumar, Vijay , month = jul, year =. doi:10.48550/arXiv.2406.17249 , abstract =

    work page doi:10.48550/arxiv.2406.17249

  25. [29]

    Distributed mapping with privacy and communication constraints:. The Int. Journal of Robotics Research , author =. 2017 , pages =. doi:10.1177/0278364917732640 , abstract =

    work page doi:10.1177/0278364917732640 2017

  26. [30]

    Cunningham, Alexander and Indelman, Vadim and Dellaert, Frank , keywords =. 2013. doi:10.1109/ICRA.2013.6631323 , abstract =

    work page doi:10.1109/icra.2013.6631323 2013

  27. [31]

    Cunningham, Alexander and Paluri, Manohar and Dellaert, Frank , keywords =. 2010. doi:10.1109/IROS.2010.5652875 , abstract =

    work page doi:10.1109/iros.2010.5652875 2010

  28. [32]

    Consensus-based

    Cristofalo, Eric and Montijano, Eduardo and Schwager, Mac , month = dec, year =. Consensus-based. 2019. doi:10.1109/CDC40024.2019.9029792 , abstract =

    work page doi:10.1109/cdc40024.2019.9029792 2019

  29. [34]

    Journal of Intelligent & Robotic Systems , author =

    Distributed. Journal of Intelligent & Robotic Systems , author =. 2025 , pages =. doi:10.1007/s10846-025-02257-w , abstract =

    work page doi:10.1007/s10846-025-02257-w 2025

  30. [35]

    and Carlone, Luca , keywords =

    Chang, Yun and Tian, Yulun and How, Jonathan P. and Carlone, Luca , keywords =. Kimera-. 2021. doi:10.1109/ICRA48506.2021.9561090 , abstract =

    work page doi:10.1109/icra48506.2021.9561090 2021

  31. [36]

    on Robotics , author =

    IEEE Trans. on Robotics , author =. 2021 , keywords =. doi:10.1109/TRO.2021.3075644 , abstract =

    work page doi:10.1109/tro.2021.3075644 2021

  32. [38]

    Field Robotics , author =

    Towards. Field Robotics , author =. 2022 , keywords =. doi:10.55417/fr.2022032 , abstract =

    work page doi:10.55417/fr.2022032 2022

  33. [39]

    IEEE Trans

    Streaming. IEEE Trans. on Signal Processing , author =. 2022 , keywords =. doi:10.1109/TSP.2022.3188208 , abstract =

    work page doi:10.1109/tsp.2022.3188208 2022

  34. [40]

    Zhang, Yetong and Hsiao, Ming and Dong, Jing and Engel, Jakob and Dellaert, Frank , keywords =. 2021. doi:10.1109/IROS51168.2021.9636687 , abstract =

    work page doi:10.1109/iros51168.2021.9636687 2021

  35. [41]

    Decentralised cooperative localisation for heterogeneous teams of mobile robots , doi =

    Bailey, Tim and Bryson, Mitch and Mu, Hua and Vial, John and McCalman, Lachlan and Durrant-Whyte, Hugh , keywords =. Decentralised cooperative localisation for heterogeneous teams of mobile robots , doi =. 2011

    work page 2011

  36. [42]

    Andersson, Lars A. A. and Nygards, Jonas , pages =. C-. 2008. doi:10.1109/ROBOT.2008.4543634 , abstract =

    work page doi:10.1109/robot.2008.4543634 2008

  37. [43]

    IEEE Trans

    Spectral. IEEE Trans. on Robotics , author =. 2024 , pages =. doi:10.1109/TRO.2023.3327635 , abstract =

    work page doi:10.1109/tro.2023.3327635 2024

  38. [44]

    2022 , keywords =

    IEEE Robotics and Automation Letters , author =. 2022 , keywords =. doi:10.1109/LRA.2022.3191204 , abstract =

    work page doi:10.1109/lra.2022.3191204 2022

  39. [45]

    Schmuck, Patrik and Ziegler, Thomas and Karrer, Marco and Perraudin, Jonathan and Chli, Margarita , month = oct, year =. 2021. doi:10.1109/ISMAR-Adjunct54149.2021.00043 , abstract =

    work page doi:10.1109/ismar-adjunct54149.2021.00043 2021

  40. [46]

    Collaborative

    Li, Fu and Yang, Shaowu and Yi, Xiaodong and Yang, Xuejun , year =. Collaborative. doi:10.1007/978-3-030-00916-8_45 , abstract =

    work page doi:10.1007/978-3-030-00916-8_45

  41. [47]

    Rotation. Int. Journal of Computer Vision , author =. 2013 , pages =. doi:10.1007/s11263-012-0601-0 , abstract =

    work page doi:10.1007/s11263-012-0601-0 2013

  42. [48]

    IEEE Trans

    Communication- and. IEEE Trans. on Robotics , author =. 2025 , pages =. doi:10.1109/TRO.2025.3567540 , abstract =

    work page doi:10.1109/tro.2025.3567540 2025

  43. [49]

    IEEE Transactions on Robotics , volume=

    Kimera-multi: Robust, distributed, dense metric-semantic slam for multi-robot systems , author=. IEEE Transactions on Robotics , volume=. 2022 , publisher=

    work page 2022

  44. [50]

    How , title =

    Aneesa Sonawalla and Yulun Tian and Jonathan P. How , title =. 2026 , journal=

    work page 2026

  45. [51]

    The invariant extended

    Barrau, Axel and Bonnabel, Silvere , journal=. The invariant extended. 2016 , publisher=

    work page 2016

  46. [52]

    borglab/gtsam , month = May, year = 2022, publisher =

    Frank Dellaert and. borglab/gtsam , month = May, year = 2022, publisher =. doi:10.5281/zenodo.5794541 , url =

    work page doi:10.5281/zenodo.5794541 2022

  47. [53]

    Agarwal, Sameer and Mierle, Keir and

    work page

  48. [55]

    Journal of Machine Learning Research , year=

    A differential equation for modeling Nesterov's accelerated gradient method: Theory and insights , author=. Journal of Machine Learning Research , year=

    work page

  49. [56]

    Soviet Mathematics Doklady , year=

    A method for solving the convex programming problem with convergence rate O (1/k2) , author=. Soviet Mathematics Doklady , year=

    work page

  50. [57]

    proceedings of the National Academy of Sciences , volume=

    A variational perspective on accelerated methods in optimization , author=. proceedings of the National Academy of Sciences , volume=. 2016 , publisher=

    work page 2016

  51. [58]

    Journal of Machine Learning Research , pages=

    A Lyapunov analysis of accelerated methods in optimization , author=. Journal of Machine Learning Research , pages=

    work page

  52. [59]

    On Symplectic Optimization

    On symplectic optimization , author=. arXiv preprint arXiv:1802.03653 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  53. [60]

    Advances in Neural Information Processing Systems , volume=

    Acceleration via symplectic discretization of high-resolution differential equations , author=. Advances in Neural Information Processing Systems , volume=

    work page

  54. [61]

    Optimization Methods and Software , volume=

    Practical perspectives on symplectic accelerated optimization , author=. Optimization Methods and Software , volume=. 2023 , publisher=

    work page 2023

  55. [62]

    Geometric numerical integration , volume=

    Structure-preserving algorithms for ordinary differential equations , author=. Geometric numerical integration , volume=. 2006 , publisher=

    work page 2006

  56. [63]

    2016 54th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , year=

    Projected gradient descent on Riemannian manifolds with applications to online power system optimization , author=. 2016 54th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , year=

    work page 2016

  57. [64]

    Journal of Nonlinear Science , volume=

    Accelerated optimization on Riemannian manifolds via discrete constrained variational integrators , author=. Journal of Nonlinear Science , volume=. 2022 , publisher=

    work page 2022

  58. [65]

    Proceedings of the IEEE conference on computer vision and pattern recognition , year=

    Gravitational approach for point set registration , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , year=

    work page

  59. [66]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Accelerated gravitational point set alignment with altered physical laws , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  60. [67]

    IEEE Transactions on Image Processing , volume=

    GraphReg: Dynamical point cloud registration with geometry-aware graph signal processing , author=. IEEE Transactions on Image Processing , volume=. 2022 , publisher=

    work page 2022

  61. [68]

    2018 International Conference on 3D Vision (3DV) , pages=

    Nrga: Gravitational approach for non-rigid point set registration , author=. 2018 International Conference on 3D Vision (3DV) , pages=. 2018 , organization=

    work page 2018

  62. [69]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Efficient registration of high-resolution feature enhanced point clouds , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2018 , publisher=

    work page 2018

  63. [70]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Dynamical pose estimation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  64. [71]

    The Euler-Poincar

    Bloch, Anthony and Krishnaprasad, PS and Marsden, Jerrold E and Ratiu, Tudor S , journal=. The Euler-Poincar. 1996 , publisher=

    work page 1996

  65. [72]

    IMA Journal of Numerical Analysis , volume=

    Global rates of convergence for nonconvex optimization on manifolds , author=. IMA Journal of Numerical Analysis , volume=. 2019 , publisher=

    work page 2019

  66. [73]

    SIAM Journal on Optimization , volume=

    Asynchronous stochastic coordinate descent: Parallelism and convergence properties , author=. SIAM Journal on Optimization , volume=. 2015 , publisher=

    work page 2015

  67. [74]

    International conference on machine learning , pages=

    Asynchronous decentralized parallel stochastic gradient descent , author=. International conference on machine learning , pages=. 2018 , organization=

    work page 2018

  68. [75]

    Georgia Institute of Technology, Tech

    Factor graphs and GTSAM: A hands-on introduction , author=. Georgia Institute of Technology, Tech. Rep , volume=

    work page

  69. [76]

    Foundations and Trends

    Factor graphs for robot perception , author=. Foundations and Trends. 2017 , publisher=

    work page 2017

  70. [77]

    Mathematical programming , volume=

    Benchmarking optimization software with performance profiles , author=. Mathematical programming , volume=. 2002 , publisher=

    work page 2002

  71. [78]

    2018 , publisher=

    Global formulations of Lagrangian and Hamiltonian dynamics on manifolds , author=. 2018 , publisher=

    work page 2018

  72. [79]

    The International Journal of Robotics Research , volume=

    Convex geometric motion planning of multi-body systems on lie groups via variational integrators and sparse moment relaxation , author=. The International Journal of Robotics Research , volume=. 2025 , publisher=

    work page 2025

  73. [80]

    IEEE Transactions on Robotics , year=

    Port-Hamiltonian neural ODE networks on Lie groups for robot dynamics learning and control , author=. IEEE Transactions on Robotics , year=

    work page

  74. [81]

    2008 , publisher=

    Optimization algorithms on matrix manifolds , author=. 2008 , publisher=

    work page 2008

  75. [82]

    Advances in neural information processing systems , volume=

    Learning to learn by gradient descent by gradient descent , author=. Advances in neural information processing systems , volume=

    work page

  76. [84]

    Proceedings of the 40th International Conference on Machine Learning , series =

    Jialin Liu and Xiaohan Chen and Zhangyang Wang and Wotao Yin and HanQin Cai , title =. Proceedings of the 40th International Conference on Machine Learning , series =

    work page

  77. [85]

    Proceedings of the 27th international conference on international conference on machine learning , pages=

    Learning fast approximations of sparse coding , author=. Proceedings of the 27th international conference on international conference on machine learning , pages=

    work page

  78. [86]

    Advances in Neural Information Processing Systems , volume=

    Accelerating quadratic optimization with reinforcement learning , author=. Advances in Neural Information Processing Systems , volume=

    work page

  79. [88]

    Saravanos and Evangelos A

    Alex Oshin and Rahul Vodeb Ghosh and Augustinos D. Saravanos and Evangelos A. Theodorou , title =. International Conference on Learning Representations , year=

    work page

  80. [89]

    IEEE Transactions on Mobile Computing , volume=

    Distributed learn-to-optimize: Limited communications optimization over networks via deep unfolded distributed ADMM , author=. IEEE Transactions on Mobile Computing , volume=. 2024 , publisher=

    work page 2024

Showing first 80 references.