Safe Bayesian Optimization for Complex Control Systems via Additive Gaussian Processes

Adrish Bhaumik; Hongxuan Wang; Lihao Zheng; Prahlad Vadakkepat; Xiaocong Li

arxiv: 2408.16307 · v3 · pith:NBH4IB35new · submitted 2024-08-29 · 💻 cs.RO · cs.AI

Safe Bayesian Optimization for Complex Control Systems via Additive Gaussian Processes

Hongxuan Wang , Xiaocong Li , Lihao Zheng , Adrish Bhaumik , Prahlad Vadakkepat This is my paper

Pith reviewed 2026-05-23 21:25 UTC · model grok-4.3

classification 💻 cs.RO cs.AI

keywords safe bayesian optimizationadditive gaussian processescontroller tuningpermanent magnet synchronous motorsafe optimizationmulti-loop controlroboticsmechatronic systems

0 comments

The pith

SafeCtrlBO uses additive Gaussian processes and a boundary-based rule to tune coupled controllers safely with fewer hardware evaluations.

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

The paper develops SafeCtrlBO to make safe Bayesian optimization practical for tuning multiple coupled controllers in robotics and mechatronics, where direct trials risk damage and existing methods demand too many expensive evaluations. It models performance and safety functions with additive Gaussian-process kernels that encode low-order interactions across controller gains, cutting the sample needs of dense high-dimensional kernels. The method swaps the costly potential-expander computation of prior SafeOpt approaches for a boundary-based expansion rule that keeps the intended safe-set growth under explicit geometric conditions. On synthetic benchmarks and a real permanent magnet synchronous motor speed-control platform, the approach reaches high-performing parameters while satisfying the high-probability safety criterion and never violating the hard signal-safety constraint.

Core claim

SafeCtrlBO simultaneously tunes multiple coupled controllers by modeling performance and constraints with additive Gaussian-process kernels that encode low-order interactions across gains. This reduces the sample complexity of full-dimensional kernels. The method substitutes the expensive potential-expander step in SafeOpt-style algorithms with a boundary-based expansion rule that maintains the safe-set expansion behavior under stated geometric conditions. On synthetic benchmarks and a PMSM speed-control platform, it attains high-performing parameters with fewer evaluations while satisfying the high-probability safety criterion and avoiding hard constraint violations.

What carries the argument

Additive Gaussian-process kernels for low-order structure across controller gains, paired with a boundary-based safe-set expansion rule.

If this is right

Reaches high-performing controller parameters with fewer hardware evaluations than representative safe BO baselines.
Maintains the prescribed high-probability safety criterion throughout optimization.
Avoids violations of the hard signal-safety constraint during hardware trials.
Enables practical tuning in moderately high-dimensional multi-loop control systems where full-dimensional kernels are prohibitive.

Where Pith is reading between the lines

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

The additive structure assumption may hold for other mechatronic plants with weakly coupled gain loops, allowing similar sample reductions without new kernel design.
The boundary rule's geometric conditions could be checked analytically on low-dimensional synthetic cases to confirm when it exactly matches the original expander behavior.
Combining the additive kernels with online kernel adaptation might further lower evaluations on plants whose interaction order changes during operation.

Load-bearing premise

The performance and constraint functions admit an additive low-order structure across controller gains that is well captured by the chosen GP kernels, and the boundary-based expansion rule preserves the intended safe-set expansion behavior under the explicit geometric conditions stated in the method.

What would settle it

On the PMSM hardware platform, if SafeCtrlBO requires at least as many evaluations as the compared safe BO baselines to reach equivalent performance or produces any violation of the hard signal-safety constraint, the claimed advantage in sample efficiency and safety preservation would not hold.

Figures

Figures reproduced from arXiv: 2408.16307 by Adrish Bhaumik, Hongxuan Wang, Lihao Zheng, Prahlad Vadakkepat, Xiaocong Li.

**Figure 2.** Figure 2: Optimization for synthetic benchmark functions. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 4.** Figure 4: Hardware experiment results. initial settings, and evaluations of these initial settings against the safety functions indicate that their values meet this minimum threshold. In the PMSM FOC control loop, the six controller parameters have different physical meanings and parameter ranges. Specifically, the proportional gain (p gain) and integral gain (i gain) of the speed controller are set within the range… view at source ↗

**Figure 5.** Figure 5: A simplified block diagram for PMSM FOC loops. The dark grey blocks represent con [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗

**Figure 6.** Figure 6: 1D visualization for the safe optimization process. The blue curve and purple shading [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗

**Figure 7.** Figure 7: 2D visualization for safe exploration. The contour values represent the predicted values of [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗

**Figure 8.** Figure 8: Ablation study using the synthetic benchmark functions. [PITH_FULL_IMAGE:figures/full_fig_p024_8.png] view at source ↗

read the original abstract

Automatic controller tuning is attractive for robotics and mechatronic systems whose dynamics are difficult to model accurately, but direct black-box optimization can be unsafe because each query is executed on the physical plant. Existing safe Bayesian optimization (BO) methods provide high-probability safety guarantees, yet their practical use in multi-loop control is limited by two coupled difficulties: the controller parameter space is often moderately high-dimensional, and hardware evaluations are too expensive to allow hundreds or thousands of exploratory trials. This paper proposes \textsc{SafeCtrlBO}, a safe BO method for simultaneously tuning multiple coupled controllers. The method uses additive Gaussian-process kernels to encode low-order structure across controller gains and reduce the sample complexity associated with dense full-dimensional kernels. It also replaces the expensive potential-expander computation used in \textsc{SafeOpt}-style exploration with a boundary-based expansion rule that preserves the intended safe-set expansion behavior under explicit geometric conditions and is validated empirically. Experiments on synthetic benchmarks and on a permanent magnet synchronous motor (PMSM) speed-control platform show that \textsc{SafeCtrlBO} reaches high-performing controller parameters with fewer hardware evaluations than representative safe BO baselines, while maintaining the prescribed high-probability safety criterion and avoiding violations of the hard signal-safety constraint in the hardware study. The code implementation is publicly available at https://github.com/hxwangnus/SafeCtrlBO.

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

SafeCtrlBO pairs additive GPs with a boundary-based safe-set rule to cut evaluations in multi-controller tuning, and the reported hardware results look usable even if the theory rests on an additive structure that may not always hold.

read the letter

The paper's core move is to replace dense GP kernels with additive ones that encode low-order interactions across controller gains, then swap the usual potential-expander step for a cheaper boundary-based expansion that keeps the safe set growing under stated geometric conditions. This targets exactly the sample-cost and dimensionality problems that have kept safe BO off real multi-loop hardware. The experiments on synthetic benchmarks and the PMSM speed-control rig show fewer evaluations than the baselines while the safety probability stays above the threshold and the hard signal constraint is never violated in the hardware runs. That combination is the concrete advance worth noting. The public code also helps anyone who wants to test it directly. The main soft spot is the additive low-order assumption itself. If the true objective or safety functions have stronger cross-term interactions than the kernels capture, the posterior used for the safety certificate becomes mis-specified and the high-probability guarantee no longer applies in the way claimed. The boundary rule is asserted to preserve the original expansion behavior only when the geometric conditions hold; the paper checks this on the reported cases, but a referee would want to see how often those conditions are actually satisfied on typical control plants and what happens when they are mildly violated. The stress-test concern therefore lands on the theory side, though the empirical safety record on the PMSM platform is still useful evidence that the method works in at least one real setting. This is aimed at people who already work on safe black-box methods for robotics or mechatronics and need something that runs on hardware without hundreds of trials. It has enough new machinery and supporting data to deserve a serious referee rather than a desk reject.

Referee Report

2 major / 1 minor

Summary. The paper proposes SafeCtrlBO, a safe Bayesian optimization algorithm for tuning multiple coupled controllers that employs additive Gaussian process kernels to capture low-order structure across gains (reducing sample complexity relative to full-dimensional kernels) and substitutes the standard potential-expander with a cheaper boundary-based safe-set expansion rule asserted to preserve the intended behavior under explicit geometric conditions. Experiments on synthetic benchmarks and a PMSM speed-control hardware platform are reported to reach high-performing parameters with fewer evaluations than safe BO baselines while satisfying the high-probability safety criterion and avoiding hard signal-safety violations.

Significance. If the additive structure is faithfully captured by the chosen kernels and the boundary rule preserves the safety certificate, the approach would meaningfully lower the number of hardware trials required for safe tuning of moderately high-dimensional multi-loop controllers, extending the practical reach of safe BO beyond the limitations of dense kernels and expensive expanders.

major comments (2)

[Method / Experiments] The high-probability safety guarantee is predicated on the objective and constraint functions admitting an additive low-order decomposition that the selected GP kernels capture with high fidelity; the manuscript provides no posterior diagnostics, kernel sensitivity analysis, or residual checks confirming this structure holds for the PMSM gains or the synthetic benchmarks (Experiments section).
[Method] The boundary-based expansion rule is stated to preserve the safe-set expansion behavior of SafeOpt-style methods only under explicit geometric conditions on the current safe set; the paper does not verify that these conditions are met on the reported benchmark and hardware instances, so the empirical absence of violations does not confirm that the theoretical certificate remains valid (Method section).

minor comments (1)

[Abstract / Introduction] The abstract and introduction would benefit from a concise statement of the precise geometric conditions required for the boundary rule.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point by point below, indicating the revisions we will make to strengthen the presentation of the safety guarantees and empirical validation.

read point-by-point responses

Referee: [Method / Experiments] The high-probability safety guarantee is predicated on the objective and constraint functions admitting an additive low-order decomposition that the selected GP kernels capture with high fidelity; the manuscript provides no posterior diagnostics, kernel sensitivity analysis, or residual checks confirming this structure holds for the PMSM gains or the synthetic benchmarks (Experiments section).

Authors: We agree that the high-probability safety claims rest on the kernels faithfully capturing the assumed additive structure. The synthetic benchmarks were constructed with explicit additive low-order interactions, and the PMSM results show practical safety, but we acknowledge the absence of explicit posterior diagnostics. In the revised manuscript we will add posterior predictive checks, kernel sensitivity analysis across length-scale and variance hyperparameters, and residual plots for both the benchmark functions and the collected PMSM data to quantify how well the additive decomposition holds. revision: yes
Referee: [Method] The boundary-based expansion rule is stated to preserve the safe-set expansion behavior of SafeOpt-style methods only under explicit geometric conditions on the current safe set; the paper does not verify that these conditions are met on the reported benchmark and hardware instances, so the empirical absence of violations does not confirm that the theoretical certificate remains valid (Method section).

Authors: The boundary-based rule is derived to match SafeOpt-style expansion precisely when the stated geometric conditions on the safe set hold. While the reported experiments exhibit no safety violations, we accept that this does not by itself confirm the conditions were satisfied throughout. In the revision we will add an explicit verification step (or a supplementary table) showing that the geometric conditions were met at each iteration for the benchmark and hardware runs, or we will clarify the precise conditions and any fallback behavior when they are not. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on explicit assumptions and external literature rather than self-referential reduction.

full rationale

The derivation introduces additive GP kernels to encode low-order controller structure and a boundary-based expansion rule asserted to preserve safe-set behavior under explicit geometric conditions. These are presented as modeling choices with empirical validation on benchmarks and hardware, not as quantities derived from or equivalent to fitted parameters or prior self-citations. The high-probability safety guarantee inherits from SafeOpt-style methods in the cited literature; no equation reduces the reported performance or safety certificate to a tautology of the paper's own inputs. Self-citations, if present, are not load-bearing for the central claims.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on standard domain assumptions from Gaussian process regression and safe optimization; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (2)

domain assumption Additive Gaussian process kernels can encode low-order structure across controller gains to reduce sample complexity relative to full-dimensional kernels.
Invoked to justify the reduction in exploratory trials for moderately high-dimensional controller spaces.
domain assumption The boundary-based expansion rule preserves high-probability safety guarantees under explicit geometric conditions.
Central to replacing the potential-expander computation while maintaining the safety property.

pith-pipeline@v0.9.0 · 5787 in / 1400 out tokens · 22802 ms · 2026-05-23T21:25:57.512158+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.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

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

additive Gaussian-process kernels to encode low-order structure across controller gains... boundary-based expansion rule... Theorem 4.1... safe set S_n = {a | l_n(a) ≥ h_i}
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

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

RKHS boundedness and Lipschitz continuity of the additive kernels... β_t = R √(2(γ_{t-1}+1+log(1/δ)))+B

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

41 extracted references · 41 canonical work pages

[1]

Relaxing the additivity constraints in decentralized no-regret high-dimensional bayesian optimization

Anthony Bardou, Patrick Thiran, and Thomas Begin. Relaxing the additivity constraints in decentralized no-regret high-dimensional bayesian optimization. In ICLR '24: International Conference on Learning Representations (ICLR), 2024

work page 2024
[2]

The curse of highly variable functions for local kernel machines

Yoshua Bengio, Olivier Delalleau, and Nicolas Roux. The curse of highly variable functions for local kernel machines. In Advances in Neural Information Processing Systems, volume 18, 2005

work page 2005
[3]

Schoellig, and Andreas Krause

Felix Berkenkamp, Angela P. Schoellig, and Andreas Krause. Safe controller optimization for quadrotors with gaussian processes. In Proc. of 2016 IEEE International Conference on Robotics and Automation (ICRA), pp.\ 491--496, Stockholm, Sweden, 2016

work page 2016
[4]

Information-theoretic safe exploration with gaussian processes

Alessandro Bottero, Carlos Luis, Julia Vinogradska, Felix Berkenkamp, and Jan R Peters. Information-theoretic safe exploration with gaussian processes. Advances in Neural Information Processing Systems, 35: 0 30707--30719, 2022

work page 2022
[5]

Haiyang Cao, Yongting Deng, Yuefei Zuo, Hongwen Li, Jianli Wang, Xiufeng Liu, and Christopher H. T. Lee. Improved adrc with a cascade extended state observer based on quasi-generalized integrator for pmsm current disturbances attenuation. IEEE Transactions on Transportation Electrification, 10 0 (1): 0 2145--2157, 2024

work page 2024
[6]

On kernelized multi-armed bandits

Sayak Ray Chowdhury and Aditya Gopalan. On kernelized multi-armed bandits. In International Conference on Machine Learning, pp.\ 844--853. PMLR, 2017

work page 2017
[7]

On kernel-target alignment

Nello Cristianini, John Shawe-Taylor, Andr\' e Elisseeff, and Jaz Kandola. On kernel-target alignment. In Advances in Neural Information Processing Systems, volume 14, 2001

work page 2001
[8]

Genetic algorithms and robotics: a heuristic strategy for optimization

Yuval Davidor. Genetic algorithms and robotics: a heuristic strategy for optimization. WORLD SCIENTIFIC, Jan. 1991

work page 1991
[9]

Approximate kernel selection via matrix approximation

Lizhong Ding, Shizhong Liao, Yong Liu, Li Liu, Fan Zhu, Yazhou Yao, Ling Shao, and Xin Gao. Approximate kernel selection via matrix approximation. IEEE Transactions on Neural Networks and Learning Systems, 31 0 (11): 0 4881--4891, 2020

work page 2020
[10]

High-dimensional gaussian process bandits

Josip Djolonga, Andreas Krause, and Volkan Cevher. High-dimensional gaussian process bandits. In Advances in Neural Information Processing Systems, volume 26, 2013

work page 2013
[11]

Additive gaussian processes

David K Duvenaud, Hannes Nickisch, and Carl Rasmussen. Additive gaussian processes. In Advances in Neural Information Processing Systems, volume 24, 2011

work page 2011
[12]

Safe contextual bayesian optimization for sustainable room temperature pid control tuning

Marcello Fiducioso, Sebastian Curi, Benedikt Schumacher, Markus Gwerder, and Andreas Krause. Safe contextual bayesian optimization for sustainable room temperature pid control tuning. In Proc. of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI-19), pp.\ 5850--5856, 2019

work page 2019
[13]

On safety in safe bayesian optimization

Christian Fiedler, Johanna Menn, Lukas Kreisk \"o ther, and Sebastian Trimpe. On safety in safe bayesian optimization. arXiv preprint arXiv:2403.12948, 2024

work page arXiv 2024
[14]

Rupprecht Gabriel, Werner Leonhard, and Craig J. Nordby. Field-oriented control of a standard ac motor using microprocessors. IEEE Transactions on Industry Applications, IA-16 0 (2): 0 186--192, 1980

work page 1980
[15]

Data-driven optimization of integrated control framework for flexible motion control system

Hanul Jung and Sehoon Oh. Data-driven optimization of integrated control framework for flexible motion control system. IEEE Transactions on Industrial Informatics, 18 0 (7): 0 4762--4772, 2022

work page 2022
[16]

High dimensional bayesian optimisation and bandits via additive models

Kirthevasan Kandasamy, Jeff Schneider, and Barnabas Poczos. High dimensional bayesian optimisation and bandits via additive models. In Proc. of the 32nd International Conference on Machine Learning (ICML), pp.\ 295--304, 2015

work page 2015
[17]

Optimizing kernel alignment over combinations of kernel

Jaz Kandola, John Shawe-Taylor, and Nello Cristianini. Optimizing kernel alignment over combinations of kernel. Technical report, Univ. Southampton Institutional Repository, Southampton, U.K., 2002

work page 2002
[18]

Smith, John Lygeros, and Alisa Rupenyan

Mohammad Khosravi, Christopher König, Markus Maier, Roy S. Smith, John Lygeros, and Alisa Rupenyan. Safety-aware cascade controller tuning using constrained bayesian optimization. IEEE Transactions on Industrial Electronics, 70 0 (2): 0 2128--2138, 2023

work page 2023
[19]

Adaptive and safe bayesian optimization in high dimensions via one-dimensional subspaces

Johannes Kirschner, Mojmir Mutny, Nicole Hiller, Rasmus Ischebeck, and Andreas Krause. Adaptive and safe bayesian optimization in high dimensions via one-dimensional subspaces. In Proc. of the 36th International Conference on Machine Learning (ICML), pp.\ 3429--3438, 2019

work page 2019
[20]

Effects of rotor position error in the performance of field-oriented-controlled pmsm drives for electric vehicle traction applications

Jorge Lara, Jianhong Xu, and Ambrish Chandra. Effects of rotor position error in the performance of field-oriented-controlled pmsm drives for electric vehicle traction applications. IEEE Transactions on Industrial Electronics, 63 0 (8): 0 4738--4751, 2016

work page 2016
[21]

Learning-based high-precision tracking control: Development, synthesis, and verification on spiral scanning with a flexure-based nanopositioner

Xiaocong Li, Haiyue Zhu, Jun Ma, Wenxin Wang, Tat Joo Teo, Chek Sing Teo, and Tong Heng Lee. Learning-based high-precision tracking control: Development, synthesis, and verification on spiral scanning with a flexure-based nanopositioner. IEEE/ASME Transactions on Mechatronics (Early Access), pp.\ 1--10, 2024

work page 2024
[22]

Bayesian approach to global optimization: theory and applications

Jonas Mockus. Bayesian approach to global optimization: theory and applications. Springer Science & Business Media, 2012

work page 2012
[23]

Efficient high dimensional bayesian optimization with additivity and quadrature fourier features

Mojmir Mutny and Andreas Krause. Efficient high dimensional bayesian optimization with additivity and quadrature fourier features. In Advances in Neural Information Processing Systems, volume 31, 2018

work page 2018
[24]

Carl Edward Rasmussen and Christopher K. I. Williams. Gaussian processes for machine learning. MIT Press, 2006

work page 2006
[25]

High-dimensional bayesian optimization via additive models with overlapping groups

Paul Rolland, Jonathan Scarlett, Ilija Bogunovic, and Volkan Cevher. High-dimensional bayesian optimization via additive models with overlapping groups. In Proc. of the Twenty-First International Conference on Artificial Intelligence and Statistics (AISTATS), pp.\ 298--307, 2018

work page 2018
[26]

Meta-learning priors for safe bayesian optimization

Jonas Rothfuss, Christopher Koenig, Alisa Rupenyan, and Andreas Krause. Meta-learning priors for safe bayesian optimization. In Proceedings of The 6th Conference on Robot Learning (CoRL), volume 205 of Proceedings of Machine Learning Research, pp.\ 237--265, 14--18 Dec 2023

work page 2023
[27]

Kakade, and Matthias Seeger

Niranjan Srinivas, Andreas Krause, Sham M. Kakade, and Matthias Seeger. Gaussian process optimization in the bandit setting: no regret and experimental design. In Proc. of the International Conference on Machine Learning (ICML), pp.\ 1015--1022, 2010

work page 2010
[28]

Safe exploration for optimization with gaussian processes

Yanan Sui, Alkis Gotovos, Joel Burdick, and Andreas Krause. Safe exploration for optimization with gaussian processes. In Proc. of the 32nd International Conference on Machine Learning (ICML), pp.\ 997--1005, Lille, France, 2015

work page 2015
[29]

Stagewise safe bayesian optimization with gaussian processes

Yanan Sui, Vincent Zhuang, Joel Burdick, and Yisong Yue. Stagewise safe bayesian optimization with gaussian processes. In Proc. of the 35th International Conference on Machine Learning (ICML), pp.\ 4781--4789, Stockholm, Sweden, 2018

work page 2018
[30]

Safe exploration for interactive machine learning

Matteo Turchetta, Felix Berkenkamp, and Andreas Krause. Safe exploration for interactive machine learning. In Advances in Neural Information Processing Systems, volume 32, 2019 a

work page 2019
[31]

Safe exploration for interactive machine learning

Matteo Turchetta, Felix Berkenkamp, and Andreas Krause. Safe exploration for interactive machine learning. Advances in Neural Information Processing Systems, 32, 2019 b

work page 2019
[32]

de Silva, and Tong Heng Lee

Wenxin Wang, Jun Ma, Zilong Cheng, Xiaocong Li, Clarence W. de Silva, and Tong Heng Lee. Global iterative sliding mode control of an industrial biaxial gantry system for contouring motion tasks. IEEE/ASME Transactions on Mechatronics, 27 0 (3): 0 1617--1628, 2022

work page 2022
[33]

de Silva, and Tong Heng Lee

Wenxin Wang, Jun Ma, Xiaocong Li, Haiyue Zhu, Clarence W. de Silva, and Tong Heng Lee. Hybrid active–passive robust control framework of a flexure-joint dual-drive gantry robot for high-precision contouring tasks. IEEE Transactions on Industrial Electronics, 70 0 (2): 0 1676--1686, 2023

work page 2023
[34]

Zheng Wang, Jian Chen, Ming Cheng, and K. T. Chau. Field-oriented control and direct torque control for paralleled vsis fed pmsm drives with variable switching frequencies. IEEE Transactions on Power Electronics, 31 0 (3): 0 2417--2428, 2016

work page 2016
[35]

On the sublinear regret of gp-ucb

Justin Whitehouse, Aaditya Ramdas, and Steven Z Wu. On the sublinear regret of gp-ucb. Advances in Neural Information Processing Systems, 36: 0 35266--35276, 2023

work page 2023
[36]

Hall, Siqi Zhou, Lukas Brunke, Melissa Greeff, Jacopo Panerati, and Angela P

Zhaocong Yuan, Adam W. Hall, Siqi Zhou, Lukas Brunke, Melissa Greeff, Jacopo Panerati, and Angela P. Schoellig. Safe-control-gym: A unified benchmark suite for safe learning-based control and reinforcement learning in robotics. IEEE Robotics and Automation Letters, 7 0 (4): 0 11142--11149, 2022

work page 2022
[37]

Robotic control optimization through kernel selection in safe bayesian optimization

Lihao Zheng, Hongxuan Wang, Xiaocong Li, Jun Ma, and Prahlad Vadakkepat. Robotic control optimization through kernel selection in safe bayesian optimization. arXiv preprint arXiv:2411.07573, 2024

work page arXiv 2024
[38]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

work page
[39]

@esa (Ref

\@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

work page
[40]

\@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

work page
[41]

@open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

work page

[1] [1]

Relaxing the additivity constraints in decentralized no-regret high-dimensional bayesian optimization

Anthony Bardou, Patrick Thiran, and Thomas Begin. Relaxing the additivity constraints in decentralized no-regret high-dimensional bayesian optimization. In ICLR '24: International Conference on Learning Representations (ICLR), 2024

work page 2024

[2] [2]

The curse of highly variable functions for local kernel machines

Yoshua Bengio, Olivier Delalleau, and Nicolas Roux. The curse of highly variable functions for local kernel machines. In Advances in Neural Information Processing Systems, volume 18, 2005

work page 2005

[3] [3]

Schoellig, and Andreas Krause

Felix Berkenkamp, Angela P. Schoellig, and Andreas Krause. Safe controller optimization for quadrotors with gaussian processes. In Proc. of 2016 IEEE International Conference on Robotics and Automation (ICRA), pp.\ 491--496, Stockholm, Sweden, 2016

work page 2016

[4] [4]

Information-theoretic safe exploration with gaussian processes

Alessandro Bottero, Carlos Luis, Julia Vinogradska, Felix Berkenkamp, and Jan R Peters. Information-theoretic safe exploration with gaussian processes. Advances in Neural Information Processing Systems, 35: 0 30707--30719, 2022

work page 2022

[5] [5]

Haiyang Cao, Yongting Deng, Yuefei Zuo, Hongwen Li, Jianli Wang, Xiufeng Liu, and Christopher H. T. Lee. Improved adrc with a cascade extended state observer based on quasi-generalized integrator for pmsm current disturbances attenuation. IEEE Transactions on Transportation Electrification, 10 0 (1): 0 2145--2157, 2024

work page 2024

[6] [6]

On kernelized multi-armed bandits

Sayak Ray Chowdhury and Aditya Gopalan. On kernelized multi-armed bandits. In International Conference on Machine Learning, pp.\ 844--853. PMLR, 2017

work page 2017

[7] [7]

On kernel-target alignment

Nello Cristianini, John Shawe-Taylor, Andr\' e Elisseeff, and Jaz Kandola. On kernel-target alignment. In Advances in Neural Information Processing Systems, volume 14, 2001

work page 2001

[8] [8]

Genetic algorithms and robotics: a heuristic strategy for optimization

Yuval Davidor. Genetic algorithms and robotics: a heuristic strategy for optimization. WORLD SCIENTIFIC, Jan. 1991

work page 1991

[9] [9]

Approximate kernel selection via matrix approximation

Lizhong Ding, Shizhong Liao, Yong Liu, Li Liu, Fan Zhu, Yazhou Yao, Ling Shao, and Xin Gao. Approximate kernel selection via matrix approximation. IEEE Transactions on Neural Networks and Learning Systems, 31 0 (11): 0 4881--4891, 2020

work page 2020

[10] [10]

High-dimensional gaussian process bandits

Josip Djolonga, Andreas Krause, and Volkan Cevher. High-dimensional gaussian process bandits. In Advances in Neural Information Processing Systems, volume 26, 2013

work page 2013

[11] [11]

Additive gaussian processes

David K Duvenaud, Hannes Nickisch, and Carl Rasmussen. Additive gaussian processes. In Advances in Neural Information Processing Systems, volume 24, 2011

work page 2011

[12] [12]

Safe contextual bayesian optimization for sustainable room temperature pid control tuning

Marcello Fiducioso, Sebastian Curi, Benedikt Schumacher, Markus Gwerder, and Andreas Krause. Safe contextual bayesian optimization for sustainable room temperature pid control tuning. In Proc. of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI-19), pp.\ 5850--5856, 2019

work page 2019

[13] [13]

On safety in safe bayesian optimization

Christian Fiedler, Johanna Menn, Lukas Kreisk \"o ther, and Sebastian Trimpe. On safety in safe bayesian optimization. arXiv preprint arXiv:2403.12948, 2024

work page arXiv 2024

[14] [14]

Rupprecht Gabriel, Werner Leonhard, and Craig J. Nordby. Field-oriented control of a standard ac motor using microprocessors. IEEE Transactions on Industry Applications, IA-16 0 (2): 0 186--192, 1980

work page 1980

[15] [15]

Data-driven optimization of integrated control framework for flexible motion control system

Hanul Jung and Sehoon Oh. Data-driven optimization of integrated control framework for flexible motion control system. IEEE Transactions on Industrial Informatics, 18 0 (7): 0 4762--4772, 2022

work page 2022

[16] [16]

High dimensional bayesian optimisation and bandits via additive models

Kirthevasan Kandasamy, Jeff Schneider, and Barnabas Poczos. High dimensional bayesian optimisation and bandits via additive models. In Proc. of the 32nd International Conference on Machine Learning (ICML), pp.\ 295--304, 2015

work page 2015

[17] [17]

Optimizing kernel alignment over combinations of kernel

Jaz Kandola, John Shawe-Taylor, and Nello Cristianini. Optimizing kernel alignment over combinations of kernel. Technical report, Univ. Southampton Institutional Repository, Southampton, U.K., 2002

work page 2002

[18] [18]

Smith, John Lygeros, and Alisa Rupenyan

Mohammad Khosravi, Christopher König, Markus Maier, Roy S. Smith, John Lygeros, and Alisa Rupenyan. Safety-aware cascade controller tuning using constrained bayesian optimization. IEEE Transactions on Industrial Electronics, 70 0 (2): 0 2128--2138, 2023

work page 2023

[19] [19]

Adaptive and safe bayesian optimization in high dimensions via one-dimensional subspaces

Johannes Kirschner, Mojmir Mutny, Nicole Hiller, Rasmus Ischebeck, and Andreas Krause. Adaptive and safe bayesian optimization in high dimensions via one-dimensional subspaces. In Proc. of the 36th International Conference on Machine Learning (ICML), pp.\ 3429--3438, 2019

work page 2019

[20] [20]

Effects of rotor position error in the performance of field-oriented-controlled pmsm drives for electric vehicle traction applications

Jorge Lara, Jianhong Xu, and Ambrish Chandra. Effects of rotor position error in the performance of field-oriented-controlled pmsm drives for electric vehicle traction applications. IEEE Transactions on Industrial Electronics, 63 0 (8): 0 4738--4751, 2016

work page 2016

[21] [21]

Learning-based high-precision tracking control: Development, synthesis, and verification on spiral scanning with a flexure-based nanopositioner

Xiaocong Li, Haiyue Zhu, Jun Ma, Wenxin Wang, Tat Joo Teo, Chek Sing Teo, and Tong Heng Lee. Learning-based high-precision tracking control: Development, synthesis, and verification on spiral scanning with a flexure-based nanopositioner. IEEE/ASME Transactions on Mechatronics (Early Access), pp.\ 1--10, 2024

work page 2024

[22] [22]

Bayesian approach to global optimization: theory and applications

Jonas Mockus. Bayesian approach to global optimization: theory and applications. Springer Science & Business Media, 2012

work page 2012

[23] [23]

Efficient high dimensional bayesian optimization with additivity and quadrature fourier features

Mojmir Mutny and Andreas Krause. Efficient high dimensional bayesian optimization with additivity and quadrature fourier features. In Advances in Neural Information Processing Systems, volume 31, 2018

work page 2018

[24] [24]

Carl Edward Rasmussen and Christopher K. I. Williams. Gaussian processes for machine learning. MIT Press, 2006

work page 2006

[25] [25]

High-dimensional bayesian optimization via additive models with overlapping groups

Paul Rolland, Jonathan Scarlett, Ilija Bogunovic, and Volkan Cevher. High-dimensional bayesian optimization via additive models with overlapping groups. In Proc. of the Twenty-First International Conference on Artificial Intelligence and Statistics (AISTATS), pp.\ 298--307, 2018

work page 2018

[26] [26]

Meta-learning priors for safe bayesian optimization

Jonas Rothfuss, Christopher Koenig, Alisa Rupenyan, and Andreas Krause. Meta-learning priors for safe bayesian optimization. In Proceedings of The 6th Conference on Robot Learning (CoRL), volume 205 of Proceedings of Machine Learning Research, pp.\ 237--265, 14--18 Dec 2023

work page 2023

[27] [27]

Kakade, and Matthias Seeger

Niranjan Srinivas, Andreas Krause, Sham M. Kakade, and Matthias Seeger. Gaussian process optimization in the bandit setting: no regret and experimental design. In Proc. of the International Conference on Machine Learning (ICML), pp.\ 1015--1022, 2010

work page 2010

[28] [28]

Safe exploration for optimization with gaussian processes

Yanan Sui, Alkis Gotovos, Joel Burdick, and Andreas Krause. Safe exploration for optimization with gaussian processes. In Proc. of the 32nd International Conference on Machine Learning (ICML), pp.\ 997--1005, Lille, France, 2015

work page 2015

[29] [29]

Stagewise safe bayesian optimization with gaussian processes

Yanan Sui, Vincent Zhuang, Joel Burdick, and Yisong Yue. Stagewise safe bayesian optimization with gaussian processes. In Proc. of the 35th International Conference on Machine Learning (ICML), pp.\ 4781--4789, Stockholm, Sweden, 2018

work page 2018

[30] [30]

Safe exploration for interactive machine learning

Matteo Turchetta, Felix Berkenkamp, and Andreas Krause. Safe exploration for interactive machine learning. In Advances in Neural Information Processing Systems, volume 32, 2019 a

work page 2019

[31] [31]

Safe exploration for interactive machine learning

Matteo Turchetta, Felix Berkenkamp, and Andreas Krause. Safe exploration for interactive machine learning. Advances in Neural Information Processing Systems, 32, 2019 b

work page 2019

[32] [32]

de Silva, and Tong Heng Lee

Wenxin Wang, Jun Ma, Zilong Cheng, Xiaocong Li, Clarence W. de Silva, and Tong Heng Lee. Global iterative sliding mode control of an industrial biaxial gantry system for contouring motion tasks. IEEE/ASME Transactions on Mechatronics, 27 0 (3): 0 1617--1628, 2022

work page 2022

[33] [33]

de Silva, and Tong Heng Lee

Wenxin Wang, Jun Ma, Xiaocong Li, Haiyue Zhu, Clarence W. de Silva, and Tong Heng Lee. Hybrid active–passive robust control framework of a flexure-joint dual-drive gantry robot for high-precision contouring tasks. IEEE Transactions on Industrial Electronics, 70 0 (2): 0 1676--1686, 2023

work page 2023

[34] [34]

Zheng Wang, Jian Chen, Ming Cheng, and K. T. Chau. Field-oriented control and direct torque control for paralleled vsis fed pmsm drives with variable switching frequencies. IEEE Transactions on Power Electronics, 31 0 (3): 0 2417--2428, 2016

work page 2016

[35] [35]

On the sublinear regret of gp-ucb

Justin Whitehouse, Aaditya Ramdas, and Steven Z Wu. On the sublinear regret of gp-ucb. Advances in Neural Information Processing Systems, 36: 0 35266--35276, 2023

work page 2023

[36] [36]

Hall, Siqi Zhou, Lukas Brunke, Melissa Greeff, Jacopo Panerati, and Angela P

Zhaocong Yuan, Adam W. Hall, Siqi Zhou, Lukas Brunke, Melissa Greeff, Jacopo Panerati, and Angela P. Schoellig. Safe-control-gym: A unified benchmark suite for safe learning-based control and reinforcement learning in robotics. IEEE Robotics and Automation Letters, 7 0 (4): 0 11142--11149, 2022

work page 2022

[37] [37]

Robotic control optimization through kernel selection in safe bayesian optimization

Lihao Zheng, Hongxuan Wang, Xiaocong Li, Jun Ma, and Prahlad Vadakkepat. Robotic control optimization through kernel selection in safe bayesian optimization. arXiv preprint arXiv:2411.07573, 2024

work page arXiv 2024

[38] [38]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

work page

[39] [39]

@esa (Ref

\@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

work page

[40] [40]

\@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

work page

[41] [41]

@open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

work page