Task-Parameter Nexus: Task-Specific Parameter Learning for Model-Based Control

Naira Hovakimyan; Ran Tao; Sheng Cheng; Shenlong Wang; Xiaofeng Wang; Yuliang Gu

REVIEW 3 major objections 4 minor 30 references

Task-Parameter Nexus: Task-Specific Parameter Learning for Model-Based Control

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A neural network can predict near-optimal control gains for unseen quadrotor tracking tasks at runtime.

desk verdict A practical method for predicting controller gains from trajectory pieces, let down by an abstract that overstates its generalization results. read the letter →

arxiv 2412.12448 v2 pith:SH5QDB74 submitted 2024-12-17 cs.RO cs.SYeess.SY

Sheng Cheng , Ran Tao , Yuliang Gu , Shenlong Wang , Xiaofeng Wang , Naira Hovakimyan This is my paper · ORCID

classification cs.ROcs.SYeess.SY

keywords optimalparameterpredictionbatchautotuningquadrotorcontroltask-parameterlearninggeometrictrajectorytrackinggeneralization

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that a model-based controller can be adapted to a new tracking task at runtime without per-task manual or offline auto-tuning, by learning the mapping from the task itself to the controller's parameters. The proposed Task-Parameter Nexus (TPN) is a neural network that takes a two-second, two-dimensional reference trajectory as input and outputs the twelve gains of a geometric quadrotor controller. The training labels come from a batch version of DiffTune that auto-tunes one parameter vector against a family of perturbed child trajectories, so the labels are robust to initial-state and waypoint variation. In simulation, the TPN's predicted gains reproduce expert tracking performance on the training categories with a maximum degradation of 4.9%, and they improve over untrained baselines on categories outside the trajectory bank and on trigonometric test trajectories.

What carries the argument

The load-bearing object is the TPN map $\varphi: \mathcal{T}_M \to \Theta$, from an $M$-step reference trajectory piece to the feasible set of controller parameters, realized as a multilayer perceptron with three hidden layers ([128, 64, 12]) and a RAYEN last layer that projects outputs into $\Theta = \{\theta \in \mathbb{R}^{12} : \theta_i \ge 0.01\}$. Around this map sit two supporting mechanisms: the trajectory bank, which arranges random minimum-snap polynomials into $3 \times 4$ speed-by-curvature categories and cuts them into two-second tasks, and batch-DiffTune, which computes expert labels by differentiating the closed-loop tracking cost through the dynamics and descending the average over child tasks with randomized initial states. The TPN is trained by mean-squared-error regression on the task-parameter pairs; the bank exists to make those supervised labels cover the motion characteristics the network must interpolate at runtime.

What would settle it

A decisive test is to take held-out trajectories with identical two-second position profiles but very different initial positions and velocities, then compare the TPN's single predicted gain vector against expert gains batch-tuned per initial-state regime: if the tracking RMSE gap between the TPN and the state-aware expert exceeds the reported 4.9% by a wide margin on such pairs, the task representation is missing a state-dependent component.

Watch

Extended reading notes

Core claim

The central claim is that the optimal control parameters of a fixed model-based controller are a learnable function of the task, where a task is encoded as a short reference trajectory. Concretely, the paper builds a trajectory bank spanning three speed levels and four curvature ranges, splits each trajectory into two-second pieces, and labels every piece with expert parameters obtained by batch-DiffTune, which minimizes the average tracking cost over randomized child trajectories and initial states. A multilayer perceptron with a constrained last layer is then trained by supervised regression to map the $201 \times 2$ position sequence to the twelve gains ($k_p$, $k_v$, $k_R$, $k_\Omega$). The finding is that this learned map matches expert parameters almost exactly on the twelve training categories, generalizes to categories outside the bank (suboptimal but far better than untrained gains), and transfers to trigonometric trajectories that never appeared in training.

Load-bearing premise

The assumption that carries the whole scheme is that a task is fully captured by a two-second, two-dimensional reference trajectory, so the optimal gains for that piece do not depend materially on the initial state or the control history that preceded it.

Editorial extensions

If this is right

If the TPN map is correct, a quadrotor can be re-tuned online every $M$ steps simply by feeding the upcoming reference piece into the network, eliminating the offline auto-tuning loop for each new task.
The bank's speed-and-curvature categories give a reusable benchmark: any controller or policy can be trained and evaluated on tasks whose translational and rotational aggressiveness are controlled independently.
Because batch-DiffTune averages over perturbed child trajectories and randomized initial states, expert labels are state-robust, so the TPN learns a task-to-parameter map rather than a state-dependent one.
On the twelve training categories the TPN stays within 4.9% of expert RMSE (S3C2), meaning the predicted gains are near-optimal for the geometric controller under randomized initial-state offsets.
TPN's gains outperform untrained parameters on categories with speeds up to 6 m/s and curvatures up to 1.2 and on circular and lemniscate trajectories, showing the learned relationship transfers beyond the training distribution.

Reading between the lines

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

A natural extension is to treat the two-second piece length $M$ as a hyperparameter: shortening it would let the TPN track rapidly changing trajectories more closely, at the cost of more frequent gain switches and a harder learning problem.
Because TPN outputs gains with no formal stability certificate beyond the empirical feasible set, one could wrap it in a safety filter that verifies a candidate gain on the current state before applying it, which the paper does not address.
The speed-curvature task parameterization likely transfers to other underactuated systems (legged robots, fixed-wing aircraft) whose tracking difficulty is dominated by the same two geometric characteristics, though the training labels would need to come from each platform's own auto-tuner.
A stress test the paper leaves implicit is to train the TPN only on low speeds and low curvatures and measure how far into the untrained high-speed, high-curvature corner its extrapolation stays useful; Table 2 suggests performance degrades as the category distance grows.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Colleague,

The TPN paper is worth knowing about: it learns a mapping from short trajectory pieces to control gains for a geometric quadrotor controller, trained on labels from their own batch-DiffTune. The core idea is simple and practical, and the trajectory bank with independent speed and curvature control is a reusable resource. The batch-DiffTune extension is a reasonable robustness fix for single-task auto-tuning.

The experimental design is mostly sound: they hold out entire categories in Table 2, randomize initial states, and report category-level statistics. The conclusion is appropriately cautious, saying TPN is suboptimal on unseen categories.

The soft spots are real, though. The abstract says 'near-optimal' and 'robust generalization to unseen tasks,' but Table 2 shows 18-27% RMSE degradation against category experts on S4C3, S5C1, S6C1. Table 1 is not a generalization test: it uses parents 17-20, which are in the training set, so it is measuring fit to training labels. The abstract's claim hangs on that table. The paper itself says 'suboptimal performance' in the conclusion, so the abstract overreaches.

Also, there is no code or data released, which limits reproducibility of the trajectory bank and the tuned labels. Table 3, the trigonometric comparison, has no expert baseline, so we cannot tell how close TPN is to optimal there. And the runtime assumption that a 2-second 2D position piece fully determines optimal gains, independent of state, is not examined empirically; the label generation averages over initial states but the TPN receives only the piece.

None of this sinks the method. It is a legitimate practical contribution with a clear scope. The claims just need to be reined in, and the authors should release code/data. I'd send it to review, expecting the authors to fix the abstract and add the missing baselines.

Referee Report

3 major / 4 minor

Summary. This paper proposes the Task-Parameter Nexus (TPN), a neural network that maps a short reference trajectory piece to the gains of a geometric controller for a quadrotor, trained in a supervised manner on labels produced by a batch version of DiffTune. The authors construct a trajectory bank spanning speed and curvature categories, generate expert parameters offline via batch-DiffTune, train a multi-layer perceptron, and evaluate tracking RMSE in simulation on training categories, held-out categories, and trigonometric trajectories. The central claim is that the TPN predicts near-optimal parameters with robust generalization to unseen tasks.

Significance. If the central claim were supported, the TPN would be a practical tool for online adaptation of model-based controller parameters to new reference trajectories without per-task auto-tuning, addressing a real limitation of low-level control. The paper's contributions include a trajectory bank with independently controlled speed and curvature, a batch extension of DiffTune that averages over randomized initial states and perturbed waypoints, and a supervised learning pipeline with a feasible-set-enforcing output layer. The experimental design with randomized initial states and multiple categories is a strength. However, the headline claim is not supported by the held-out results: the in-distribution table is a training-set fit, and the only held-out categories show 17-27% degradation relative to the batch-DiffTune expert. As a result, the paper currently overstates the empirical case for near-optimal robust generalization.

major comments (3)

[Section 5 (Training of the TPN; Table 1)] The TPN is trained on all 1200 parent-piece pairs, i.e., 12 categories × 20 parents × 5 pieces, yet Table 1 evaluates the TPN on trajectory piece 1 of parents 17-20, which are among those 20 parents. Consequently, Table 1 measures fit to the training labels, not generalization to unseen tasks, and the 'maximum degradation of 4.9%' cannot support the abstract's 'near-optimal ... robust generalization to unseen tasks.' The only genuinely held-out comparison, Table 2, shows TPN RMSE exceeding expert RMSE by 18% in S4C3 (0.273 vs 0.232), 17% in S5C1 (0.244 vs 0.208), and 27% in S6C1 (0.263 vs 0.207), which the conclusion itself describes as 'suboptimal.' I recommend either tempering the abstract and Section 1 claims to match this evidence or adding a proper held-out evaluation, e.g., training on 16 parents and testing on 4 held-out parents within the same categories, to demonstrate in-distribution generalization.
[Section 4.2 (Batch-DiffTune, Eq. (PO) and (7))] The 'expert parameters' are the result of batch gradient descent on the authors' own batch-DiffTune objective with a fixed step size and 100 iterations. The term 'near-optimal' in the abstract is therefore circular: the TPN is trained to match the outputs of the authors' optimizer, and the evaluation compares against the same optimizer. To substantiate the optimality claim, the paper should compare against an independent tuning baseline (e.g., grid search, Bayesian optimization, or a different auto-tuning method) or report the gap to a more exhaustive optimization. Absent such a baseline, the claim that the TPN produces 'near-optimal' parameters is unsupported, though the downstream comparison still shows the TPN improves over untrained parameters.
[Section 3 and Section 4.2 (Task representation and state-independence assumption)] The TPN input is only a 2-second, 2D position trajectory piece, while the batch-DiffTune labels are averaged over randomized initial states and child waypoint perturbations. The paper assumes the optimal parameters are well-defined functions of the trajectory piece alone (state-independent and piece-local), but this assumption is not tested. At runtime, the TPN receives no initial state, velocity, or neighboring-piece context, and the loop uses piecewise-constant parameters updated every M steps. If the optimal gains depend on the tracking error state or on the longer-horizon trajectory, the piecewise state-free prediction cannot be near-optimal, especially in the aggressive S3C4 regime. I suggest an ablation that conditions the TPN on the current state (or on a longer trajectory window) and checks whether the held-out RMSE improves; this would test the adequacy of the task representation.

minor comments (4)

[Abstract and Section 6] The abstract's 'near-optimal ... robust generalization capabilities to unseen tasks' is stronger than the conclusion's 'suboptimal performance on the tasks within unseen categories'; please align the abstract with the empirical evidence.
[Section 4.3] The sentence 'It can infer parameter choice for unseen tasks (trajectories) whose motion characteristics have already been included in the trajectory bank and, thus, in the training set of the TPN' is internally contradictory because tasks whose characteristics are in the training set are not unseen; please clarify what 'unseen' means (e.g., new parametrizations within the same category).
[Table 3 caption] The word 'Leminiscate' should be spelled 'Lemniscate'.
[Section 5 (TPN training)] The text says training is completed in '50 episodes'; this is likely 'epochs'. Also, the paper does not state how the 1200 pairs are split into training and validation sets for the TPN, despite reporting validation loss; please specify the split.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: TPN is a supervised fit to auto-tuned labels, and the central performance claims are checked by independent closed-loop simulation; the unsupported generalization claim is an overstatement, not circularity.

full rationale

The claimed derivation is linear: trajectories are generated (Section 4.1), expert parameters are produced by batch-DiffTune as the minimizer of (PO) (Section 4.2), the TPN is trained to fit those labels via MSE (8), and performance is measured by closed-loop RMSE in simulation (Section 5). None of these steps defines the TPN output in terms of the reported tracking error, nor does any equation equate prediction with input. The labels inherit the quality of batch-DiffTune, which is the authors' own extension of DiffTune (Cheng et al., 2024), but the evaluation of TPN-vs-expert RMSE is an independent dynamical simulation with randomized initial states, so the central claim is not forced by construction. The abstract's 'robust generalization to unseen tasks' is not supported by Table 1, which evaluates trajectory piece 1 of parents 17-20, all of which are included in the 1200 training pairs; the genuinely held-out categories in Table 2 show 18-27% degradation. That is a correctness/overclaim concern, not a circularity concern. No self-citation is load-bearing in the sense of a uniqueness theorem or an unverified ansatz; DiffTune is a published, reproducible optimization method whose outputs are externally falsifiable. Score 0.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

No new physical entities are introduced. The paper's central claim rests on the representativeness of the speed-curvature trajectory bank, the quality of batch-DiffTune labels as a proxy for optimality, and the assumption that a short position trajectory fully determines the controller gains. The free parameters listed are the hand-chosen design values that set the data distribution and training procedure.

free parameters (8)

Speed interval medians = 1, 2, 3 m/s
Define the three speed categories S1-S3; the trajectory bank and the TPN training distribution depend on these values.
Curvature interval boundaries = [0,0.2], [0.2,0.4], [0.4,0.6], [0.6,0.8] 1/m
Define curvature categories C1-C4; affects task difficulty coverage and generalization behavior.
Child perturbation radius r = 0.05 m
Sets the domain randomization magnitude in batch-DiffTune; too large would change the category, too small reduces robustness.
Batch-DiffTune step size alpha = 0.1
Controls gradient descent convergence; chosen by hand without a sensitivity study.
Batch-DiffTune iterations = 100
Stop criterion for the parameter update; no convergence analysis is provided.
DNN hyperparameters = hidden [128,64,12], lr 0.001, batch 32, epochs 50
Architecture and training choices are reported but no ablations are shown.
Feasible-set margin = 0.01
The output layer enforces theta >= 0.01; the margin is chosen to keep parameters positive, with no analysis of its effect.
Initial parameter values for auto-tuning = kp=16, kv=5.6, kR=8.81, kOmega=2.54
Initial guess for batch-DiffTune; results may depend on this seed for local optima.

assumptions (6)

domain assumption Quadrotor dynamics (9) and the SE(3) geometric controller (10)-(11) accurately model the closed-loop system.
The entire simulation and auto-tuning rely on these models; no model mismatch is considered, so the tuned parameters are optimal for this model, not for a real quadrotor.
domain assumption The speed and curvature of a trajectory are sufficient coordinates to characterize the difficulty of a tracking task.
Section 4.1 constructs the task space using only speed and curvature; higher-order effects such as jerk, snap, and yaw profile are not independently represented in the category labels.
domain assumption The optimal parameter for a task is well-defined and independent of initial state and control history.
Section 4.2 averages over randomized initial states to make expert parameters depend on tasks only; the TPN itself receives only the trajectory piece.
domain assumption Batch-DiffTune converges to a locally optimal parameter that is a suitable ground truth for training.
The paper uses gradient descent with 100 fixed iterations and no certificate of global optimality.
standard math Differential flatness and minimum-snap trajectory generation can produce any smooth reference needed for tasks.
Used in Section 4.1 to convert polynomials into full-state reference trajectories; this is a standard and accepted result for quadrotors.
domain assumption The empirical feasible set Theta = {theta | theta >= 0.01} guarantees stability.
Section 4.3 uses this set with no proof; the paper notes stability can be guaranteed analytically or empirically, but no analysis is given for this specific margin.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Task-Parameter Nexus: Task-Specific Parameter Learning for Model-Based Control." pith.science (2026). https://pith.science/paper/SH5QDB74

@misc{pith2026241212448,
  author       = {Pith},
  title        = {Pith review of: Task-Parameter Nexus: Task-Specific Parameter Learning for Model-Based Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SH5QDB74}},
  note         = {Machine review of arXiv:2412.12448}
}

read the original abstract

This paper presents the Task-Parameter Nexus (TPN), a learning-based approach for online determination of the (near-)optimal control parameters of model-based controllers (MBCs) for tracking tasks. In TPN, a deep neural network is introduced to predict the control parameters for any given tracking task at runtime, especially when optimal parameters for new tasks are not immediately available. To train this network, we constructed a trajectory bank with various speeds and curvatures that represent different motion characteristics. Then, for each trajectory in the bank, we auto-tune the optimal control parameters offline and use them as the corresponding ground truth. With this dataset, the TPN is trained by supervised learning. We evaluated the TPN on the quadrotor platform. In simulation experiments, it is shown that the TPN can predict near-optimal control parameters for a spectrum of tracking tasks, demonstrating its robust generalization capabilities to unseen tasks.

Figures

Figures reproduced from arXiv: 2412.12448 by the authors.

**Figure 2.** Illustration of the trajectory bank (consisting of trajectories under the hierarchy of [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Box plots of speed (a) and curvature (b) over the 12 categories in the trajectory bank. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Trajectories ran by expert parameters (blue) vs untrained parameters (red) on three training tasks (black). Only three trajectories are shown for illustration purposes. Subfigures from left to right, top to bottom are from S1C1, S1C4, S3C1, S3C4. conduct training for…

**Figure 5.** Figure 5: Illustration of parent 1 over all 12 categories. Each segment in between two dots has [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]

**Figure 7.** Figure 7: Evaluation of the expert parameters in the testing tasks shown in a box plot. The vertical axis refers to the ratio of mean RMSE achieved by expert parameters on the training tasks divided by those on the testing tasks. To validate batch-tuned expert parameters, we eva…

**Figure 6.** Figure 6: Illustration of parent trajectory and randomized child trajectories surrounding the parent. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 23 canonical work pages

[1]

Safe controller optimization for quadrotors with Gaussian processes

Felix Berkenkamp, Angela P Schoellig, and Andreas Krause. Safe controller optimization for quadrotors with Gaussian processes . In Proceedings of IEEE International Conference on Robotics and Automation, pages 491--496, Stockholm, Sweden, 2016

work page 2016
[2]

Imitate and repurpose: Learning reusable robot movement skills from human and animal behaviors

Steven Bohez, Saran Tunyasuvunakool, Philemon Brakel, Fereshteh Sadeghi, Leonard Hasenclever, Yuval Tassa, Emilio Parisotto, Jan Humplik, Tuomas Haarnoja, Roland Hafner, et al. Imitate and repurpose: Learning reusable robot movement skills from human and animal behaviors. arXiv preprint arXiv:2203.17138, 2022

arXiv 2022
[3]

RT-2 : Vision-language-action models transfer web knowledge to robotic control

Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, et al. RT-2 : Vision-language-action models transfer web knowledge to robotic control. arXiv preprint arXiv:2307.15818, 2023

arXiv 2023
[4]

Safe learning in robotics: From learning-based control to safe reinforcement learning

Lukas Brunke, Melissa Greeff, Adam W Hall, Zhaocong Yuan, Siqi Zhou, Jacopo Panerati, and Angela P Schoellig. Safe learning in robotics: From learning-based control to safe reinforcement learning. Annual Review of Control, Robotics, and Autonomous Systems, 5: 0 411--444, 2022

work page 2022
[5]

DiffTune +: Hyperparameter-free auto-tuning using auto-differentiation

Sheng Cheng, Lin Song, Minkyung Kim, Shenlong Wang, and Naira Hovakimyan. DiffTune +: Hyperparameter-free auto-tuning using auto-differentiation. In Learning for Dynamics and Control Conference, pages 170--183. PMLR, 2023

work page 2023
[6]

DiffTune : Auto-tuning through auto-differentiation

Sheng Cheng, Minkyung Kim*, Lin Song*, Chengyu Yang, Yiquan Jin, Shenlong Wang, and Naira Hovakimyan. DiffTune : Auto-tuning through auto-differentiation. IEEE Transactions on Robotics, 2024

work page 2024
[7]

Online safe trajectory generation for quadrotors using fast marching method and B ernstein basis polynomial

Fei Gao, William Wu, Yi Lin, and Shaojie Shen. Online safe trajectory generation for quadrotors using fast marching method and B ernstein basis polynomial. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 344--351. IEEE, 2018

work page 2018
[8]

Learning agile and dynamic motor skills for legged robots

Jemin Hwangbo, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter. Learning agile and dynamic motor skills for legged robots. Science Robotics, 4 0 (26): 0 eaau5872, 2019

2019

Show all 30 references

[9]

DiffLoop : Tuning PID controllers by differentiating through the feedback loop

Athindran Ramesh Kumar and Peter J Ramadge. DiffLoop : Tuning PID controllers by differentiating through the feedback loop. In Proceedings of the 55th Annual Conference on Information Sciences and Systems, pages 1--6, Baltimore, MD, USA, 2021

2021
[10]

Geometric tracking control of a quadrotor UAV on SE (3)

Taeyoung Lee, Melvin Leok, and N Harris McClamroch. Geometric tracking control of a quadrotor UAV on SE (3). In Proceedings of the 49th IEEE Conference on Decision and Control, pages 5420--5425, Atlanta, GA, USA, 2010 a

2010
[11]

Control of complex maneuvers for quadrotor UAV using geometric methods on SE (3)

Taeyoung Lee, Melvin Leok, and N Harris McClamroch. Control of complex maneuvers for quadrotor UAV using geometric methods on SE (3). arXiv:1003.2005, 2010 b

2005 arXiv
[12]

Improved baselines with visual instruction tuning

Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning. In Proceedings of International Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2024

2024
[13]

Autotune: Controller tuning for high-speed flight

Antonio Loquercio, Alessandro Saviolo, and Davide Scaramuzza. Autotune: Controller tuning for high-speed flight. IEEE Robotics and Automation Letters, 7 0 (2): 0 4432--4439, 2022

2022
[14]

DeepSeek-VL : Towards real-world vision-language understanding, 2024

Haoyu Lu, Wen Liu, Bo Zhang, Bingxuan Wang, Kai Dong, Bo Liu, Jingxiang Sun, Tongzheng Ren, Zhuoshu Li, Yaofeng Sun, Chengqi Deng, Hanwei Xu, Zhenda Xie, and Chong Ruan. DeepSeek-VL : Towards real-world vision-language understanding, 2024

2024
[15]

Minimum snap trajectory generation and control for quadrotors

Daniel Mellinger and Vijay Kumar. Minimum snap trajectory generation and control for quadrotors. In Proceedings fo the International Conference on Robotics and Automation, pages 2520--2525, Shanghai, China, 2011

2011
[16]

Untersuchungen \"u ber allgemeine metrik

Karl Menger. Untersuchungen \"u ber allgemeine metrik. Mathematische Annalen, 103 0 (1): 0 466--501, 1930

1930
[17]

Linearizing large language models

Jean Mercat, Igor Vasiljevic, Sedrick Keh, Kushal Arora, Achal Dave, Adrien Gaidon, and Thomas Kollar. Linearizing large language models. arXiv preprint arXiv:2405.06640, 2024

2024 arXiv
[18]

Local policy search with B ayesian optimization

Sarah M \"u ller, Alexander von Rohr, and Sebastian Trimpe. Local policy search with B ayesian optimization. In Advances in Neural Information Processing Systems, 2021

2021
[19]

Research on gain scheduling

Wilson J Rugh and Jeff S Shamma. Research on gain scheduling. Automatica, 36 0 (10): 0 1401--1425, 2000

2000
[20]

Physics-inspired temporal learning of quadrotor dynamics for accurate model predictive trajectory tracking

Alessandro Saviolo, Guanrui Li, and Giuseppe Loianno. Physics-inspired temporal learning of quadrotor dynamics for accurate model predictive trajectory tracking. IEEE Robotics and Automation Letters, 7 0 (4): 0 10256--10263, 2022

2022
[21]

Policy search for model predictive control with application to agile drone flight

Yunlong Song and Davide Scaramuzza. Policy search for model predictive control with application to agile drone flight. IEEE Transactions on Robotics, 38 0 (4): 0 2114--2130, 2022

2022
[22]

A comparative study of nonlinear MPC and differential-flatness-based control for quadrotor agile flight

Sihao Sun, Angel Romero, Philipp Foehn, Elia Kaufmann, and Davide Scaramuzza. A comparative study of nonlinear MPC and differential-flatness-based control for quadrotor agile flight. IEEE Transactions on Robotics, 38 0 (6): 0 3357--3373, 2022

2022
[23]

DiffTune-MPC : Closed-loop learning for model predictive control

Ran Tao, Sheng Cheng, Xiaofeng Wang, Shenlong Wang, and Naira Hovakimyan. DiffTune-MPC : Closed-loop learning for model predictive control. IEEE Robotics and Automation Letters, 2024

2024
[24]

RAYEN : Imposition of hard convex constraints on neural networks

Jesus Tordesillas, Jonathan P How, and Marco Hutter. RAYEN : Imposition of hard convex constraints on neural networks. arXiv preprint arXiv:2307.08336, 2023

2023 arXiv
[25]

Data-driven MPC for quadrotors

Guillem Torrente, Elia Kaufmann, Philipp F \"o hn, and Davide Scaramuzza. Data-driven MPC for quadrotors. IEEE Robotics and Automation Letters, 6 0 (2): 0 3769--3776, 2021

2021
[26]

Llama: Open and efficient foundation language models

Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth \'e e Lacroix, Baptiste Rozi \`e re, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023

2023 arXiv
[27]

Vision LLM : Large language model is also an open-ended decoder for vision-centric tasks

Wenhai Wang, Zhe Chen, Xiaokang Chen, Jiannan Wu, Xizhou Zhu, Gang Zeng, Ping Luo, Tong Lu, Jie Zhou, Yu Qiao, et al. Vision LLM : Large language model is also an open-ended decoder for vision-centric tasks. Advances in Neural Information Processing Systems, 36, 2024

2024
[28]

L _1 adaptive augmentation for geometric tracking control of quadrotors

Zhuohuan Wu, Sheng Cheng, Kasey A Ackerman, Aditya Gahlawat, Arun Lakshmanan, Pan Zhao, and Naira Hovakimyan. L _1 adaptive augmentation for geometric tracking control of quadrotors. In Proceedings of the International Conference on Robotics and Automation, pages 1329--1336, P...

2022
[29]

L _1 Quad : L _1 adaptive augmentation of geometric control for agile quadrotors with performance guarantees

Zhuohuan Wu, Sheng Cheng, Pan Zhao, Aditya Gahlawat, Kasey A Ackerman, Arun Lakshmanan, Chengyu Yang, Jiahao Yu, and Naira Hovakimyan. L _1 Quad : L _1 adaptive augmentation of geometric control for agile quadrotors with performance guarantees. arXiv preprint arXiv:2302.07208, 2023

2023 arXiv
[30]

Robust and efficient quadrotor trajectory generation for fast autonomous flight

Boyu Zhou, Fei Gao, Luqi Wang, Chuhao Liu, and Shaojie Shen. Robust and efficient quadrotor trajectory generation for fast autonomous flight. IEEE Robotics and Automation Letters, 4 0 (4): 0 3529--3536, 2019

2019

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Safe controller optimization for quadrotors with Gaussian processes

Felix Berkenkamp, Angela P Schoellig, and Andreas Krause. Safe controller optimization for quadrotors with Gaussian processes . In Proceedings of IEEE International Conference on Robotics and Automation, pages 491--496, Stockholm, Sweden, 2016

work page 2016

[2] [2]

Imitate and repurpose: Learning reusable robot movement skills from human and animal behaviors

Steven Bohez, Saran Tunyasuvunakool, Philemon Brakel, Fereshteh Sadeghi, Leonard Hasenclever, Yuval Tassa, Emilio Parisotto, Jan Humplik, Tuomas Haarnoja, Roland Hafner, et al. Imitate and repurpose: Learning reusable robot movement skills from human and animal behaviors. arXiv preprint arXiv:2203.17138, 2022

arXiv 2022

[3] [3]

RT-2 : Vision-language-action models transfer web knowledge to robotic control

Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, et al. RT-2 : Vision-language-action models transfer web knowledge to robotic control. arXiv preprint arXiv:2307.15818, 2023

arXiv 2023

[4] [4]

Safe learning in robotics: From learning-based control to safe reinforcement learning

Lukas Brunke, Melissa Greeff, Adam W Hall, Zhaocong Yuan, Siqi Zhou, Jacopo Panerati, and Angela P Schoellig. Safe learning in robotics: From learning-based control to safe reinforcement learning. Annual Review of Control, Robotics, and Autonomous Systems, 5: 0 411--444, 2022

work page 2022

[5] [5]

DiffTune +: Hyperparameter-free auto-tuning using auto-differentiation

Sheng Cheng, Lin Song, Minkyung Kim, Shenlong Wang, and Naira Hovakimyan. DiffTune +: Hyperparameter-free auto-tuning using auto-differentiation. In Learning for Dynamics and Control Conference, pages 170--183. PMLR, 2023

work page 2023

[6] [6]

DiffTune : Auto-tuning through auto-differentiation

Sheng Cheng, Minkyung Kim*, Lin Song*, Chengyu Yang, Yiquan Jin, Shenlong Wang, and Naira Hovakimyan. DiffTune : Auto-tuning through auto-differentiation. IEEE Transactions on Robotics, 2024

work page 2024

[7] [7]

Online safe trajectory generation for quadrotors using fast marching method and B ernstein basis polynomial

Fei Gao, William Wu, Yi Lin, and Shaojie Shen. Online safe trajectory generation for quadrotors using fast marching method and B ernstein basis polynomial. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 344--351. IEEE, 2018

work page 2018

[8] [8]

Learning agile and dynamic motor skills for legged robots

Jemin Hwangbo, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter. Learning agile and dynamic motor skills for legged robots. Science Robotics, 4 0 (26): 0 eaau5872, 2019

2019