ACORN: Adaptive Contrastive Optimization for Safe and Robust Fine-Grained Robotic Manipulation

Shuhao Li; Zhongquan Zhou; Zixian Yue

REVIEW 4 major objections 5 minor 29 references

ACORN: Adaptive Contrastive Optimization for Safe and Robust Fine-Grained Robotic Manipulation

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A new loss makes robot imitation policies safer under noise, lifting failure-conditional reward by 23.1% without hurting success rate.

desk verdict A plausible contrastive robustness add-on for ACT, but the headline ACR-F metric depends on an undefined reward and the evaluation lacks error bars, so the central claim needs better support before it can be trusted. read the letter →

arxiv 2505.06628 v1 pith:UVGT572J submitted 2025-05-10 cs.RO

Zhongquan Zhou , Shuhao Li , Zixian Yue This is my paper · ORCID

classification cs.RO

keywords roboticmanipulationimitationlearningActionChunkingwithTransformerscontrastivesafetymetricsperturbationrobustnesscurriculumACORN

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

ACORN claims to be a plug-and-play training objective that extends Action Chunking with Transformers (ACT), a common imitation learning method for fine-grained bimanual manipulation. It adds a contrastive loss that pulls predicted action chunks toward expert demonstrations and pushes them away from synthetically perturbed 'unsafe' action chunks, with a curriculum schedule that grows the contrastive weight as the baseline loss drops. The paper reports that under a normal level of actuator noise, ACT-ACORN matches the baseline success rate (62.2%) while raising the failure-conditional reward ACR-F from 69.95 to 86.12, a 23.1% improvement. If true, ACORN gives a drop-in robustness enhancement that requires no extra demonstrations and no changes to the policy architecture.

What carries the argument

The central object is the ACT-ACORN loss, $$L_{\text{ACT-ACORN}} = L_{\text{Huber}} + \lambda_{\text{KL}} L_{\text{KL}}^D + \lambda_c(L_b) L_{\text{Contrast}},$$ with $$L_{\text{Contrast}} = \max(0, \|a^+ - \hat{a}\|_2 - \|a^- - \hat{a}\|_2 + \$\alpha$).$$ Negative action chunks are generated by dual perturbation, $$a^- = (1 + \eta\$\sigma$) \odot a_{gt} + \epsilon,$$ combining correlated scaling with independent micro-perturbations. The contrastive weight $\lambda_c(L_b)$ follows a curriculum schedule driven by the baseline loss, so the policy first learns to imitate and then increasingly learns to reject noisy trajectories. This machinery turns every training batch into paired positive and negative action examples, teaching the policy to hug expert demonstrations while actively avoiding unsafe actions.

What would settle it

Replicate the AlohaTransferCube comparison and require the authors to specify $r_t$ and compute ACR-F explicitly; if no reward exists in the imitation-learning setup, the claimed 23.1% gap cannot be reproduced. A simpler check is to rerun the normal-noise evaluation with additional random seeds and see whether the ACR-F difference between ACT and ACT-ACORN persists.

Watch

Extended reading notes

Core claim

The paper's central claim is that ACORN, an extension of ACT, improves safety and robustness against actuator noise without sacrificing task performance. The mechanism is a composite objective in which the usual ACT losses (an adaptive Huber regression loss and KL regularization) are augmented by a contrastive margin loss that pushes predicted action chunks toward expert demonstrations and away from synthetically corrupted ones. Negative samples are generated by correlated Gaussian scaling plus tiny independent perturbations, and the contrastive weight grows through a curriculum schedule tied to the baseline loss. In the reported AlohaTransferCube evaluation, ACT-ACORN matches the baseline success rates under all tested noise levels (80.80 vs 77.40 for light, 62.20 vs 62.20 for normal, 22.80 vs 23.00 for heavy), while substantially improving ACR-F under normal and heavy noise and reducing AM-E, which the paper interprets as safer, more precise, and more energy-efficient manipulation.

Load-bearing premise

The load-bearing premise is that the headline safety metric ACR-F is computable, but the paper never defines the reward function $r_t$ that ACR-F sums, and the supervised imitation-learning setup supplies no reward signal, so the 23.1% improvement may rest on an uncomputed quantity.

Editorial extensions

If this is right

Any ACT-style imitation policy could be hardened against action noise by adding dual-perturbation negative samples and a curriculum-scheduled contrastive loss, with no new demonstrations or network changes.
If the reported equal success rate under normal noise holds, robustness improvements need not trade off task completion, at least on the AlohaTransferCube task.
The adaptive Huber loss and the margin parameter provide new tuning knobs whose effects on safety metrics are demonstrated in the ablations.
The four safety metrics, once the reward component is defined, could become a standard way to compare manipulation policies under disturbance.
The biomechanical joint grouping suggests that concentrating contrastive weight on high-priority joints could be a cheap way to target precision improvements, though the paper only visualizes this effect and does not ablate it separately.

Reading between the lines

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

Because the contrastive loss operates purely on action chunks, the same negative-sample recipe should transfer to other action-sequence policies beyond ACT, though this is not tested in the paper.
The joint grouping (shoulder, elbow, and forearm roll versus waist and wrist) suggests a testable extension: apply larger contrastive weights to high-priority joints and measure whether safety gains concentrate there.
Until a reward function for ACR-F is specified, AM-J and AM-E are the only directly computable safety metrics in the paper, so reproducing the headline 23.1% improvement requires that missing definition.
The curriculum schedule, with its sharp transition once the baseline loss drops below one, invites sensitivity tests on other tasks, since the exponential coefficient $k$ is tuned on a single environment.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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,

ACORN is a contrastive-learning add-on to ACT that tries to make imitation policies robust to action noise. The core idea is reasonable: generate noisy negative action trajectories, then push the policy away from them with a contrastive loss while keeping it close to expert actions. The paper also adds a curriculum schedule for the contrastive weight and a Huber loss. That is a sensible combination, and the ablation studies for k and α are actually informative.

The problem is that the headline result—a 23.1% improvement in ACR-F under Normal noise—is not something you can check. ACR-F, defined in Eq (3), is an average cumulative reward over failed episodes. But no reward function r_t is defined anywhere. The task is supervised imitation learning on AlohaTransferCube; there is no reward signal in that setup. So the 86.12 vs 69.95 number is uninterpretable. The same goes for ACR in Table I. This is not a minor omission; it is load-bearing. Without a definition of r_t, the direction of the comparison is unconstrained. A reader cannot verify that ACORN is safer rather than just gaming an unstated scoring rule.

The other metrics are also weaker than they look. AM-J and AM-E measure deviation from the training demonstrations, which is partly by construction: ACORN is trained to imitate those demonstrations, so lower values are somewhat expected. The evaluation injects multiplicative Gaussian action noise (Eq 6), the same family used to generate the negative samples (Eq 7), which raises a question about overfitting to that noise profile. There are no error bars, no multiple seeds, only one task and one baseline. And the hyperparameters (k=15, α=0.01) were selected via ablation on the Normal noise condition that also appears in the main results—a mild test-set tuning issue.

What the paper does well: the method is simple and plausible, the presentation is clear, and the authors are upfront about the architecture and training details. The idea of using contrastive learning to separate expert and noisy action trajectories is worth exploring. But the evidence, as it stands, does not support the claim.

For peer review: I would send it to referees rather than desk reject, because the core idea has merit and the flaws are fixable. But I would reject in current form and ask for a defined reward, error bars, an additional task, and an evaluation that distinguishes safety from closeness to the training set.

That's my take. Let's talk over coffee if you want.

Referee Report

4 major / 5 minor

Summary. The paper proposes ACORN, a plug-and-play extension of the Action Chunking with Transformers (ACT) imitation learning policy. ACORN adds contrastive negative sampling via correlated scaling and independent micro-perturbations, a curriculum-scheduled contrastive loss, and a Huber regression term. The paper also introduces four safety metrics (ACR-F, AM-J, AM-E, TDL) and reports experiments on the LeRobot AlohaTransferCube task under Light, Normal, and Heavy multiplicative Gaussian action noise, claiming up to 23% improvements in ACR-F without sacrificing success rate.

Significance. If the claims held, ACORN would be a practically valuable robustness wrapper for ACT-style policies, since it is architecture-compatible and requires no additional data collection. The paper also contributes a plausible evaluation template that perturbs actions at three noise levels, and it reports ablations for the curriculum coefficient k and the contrastive margin alpha. However, the headline claim depends on an undefined reward function, the evaluation protocol overlaps with the training-time noise family, and no variance or multi-seed statistics are reported. The central safety claim is therefore not currently supported, and the contribution remains an idea plus unverified numbers rather than a validated method.

major comments (4)

[Section IV-A, Eq. (3); Table I] ACR-F is defined as the average cumulative reward over failed episodes, but no reward function r_t^(i) is defined anywhere in the manuscript. The ACT objective in Eq. (1) is supervised imitation (L1 reconstruction plus KL regularization), and the LeRobot AlohaTransferCube task is an imitation-learning benchmark with no reward signal. Consequently, the Normal-noise ACR-F comparison in Table I (86.116 vs 69.952) and the abstract's 23% improvement claim cannot be computed or independently verified. This is load-bearing because ACR and ACR-F are used as primary evidence of safety improvement; the paper must either define the reward used in evaluation or remove reward-based metrics.
[Section V-A, Table I] No variance, confidence intervals, or multiple seeds are reported. The text states that 500 episodes per noise condition were evaluated, but each entry in Table I is a single point estimate. Given the stochasticity of the noise injection in Eq. (6) and the high variance of manipulation success rates (SR ranges from 22.8% to 80.8% across conditions), small differences such as SR 22.80 vs 23.00 or AM-E 0.116 vs 0.115 are indistinguishable from sampling noise. To support the claimed improvements, the authors should report mean and standard deviation over at least 3-5 independent training or evaluation seeds.
[Section IV-B, Eq. (6); Section IV-C, Eq. (7)] The evaluation noise family matches the training-time negative sample distribution. Eq. (6) generates perturbed actions as (1+beta) a_gt with beta = eta*sigma and eta ~ N(0,1), i.e., multiplicative Gaussian scaling. Eq. (7) generates negative samples as (1 + eta*sigma) * a_gt + epsilon, which is the same multiplicative Gaussian family plus an independent micro-perturbation. ACORN is therefore evaluated on the perturbation distribution it was explicitly trained against, while the ACT baseline was not exposed to this distribution. This confounds the robustness comparison and does not demonstrate generalization to realistic disturbances. A concrete fix is to evaluate on held-out noise families (additive Gaussian, uniform, sensor noise) or to train the baseline with the same negative-sample augmentation.
[Section IV-A, Eqs. (4)-(5); Section V-B] AM-J and AM-E measure deviation from the nearest training demonstrations, and ACORN's contrastive loss explicitly minimizes distance to expert positive samples while ACT only regresses with L1/Huber loss. Lower AM-J/AM-E under noise is therefore partly expected by construction and does not, by itself, establish safer behavior. The paper should validate these kinematic metrics against independent safety indicators, such as workspace-boundary or obstacle violations, and should report TDL quantitatively rather than only as a visualization concept, since Table I contains no TDL values.

minor comments (5)

[Section IV, Stage II] The text contains a duplicated phrase: 'with with 1) Huber Loss and 2) Contrastive Margin Loss'.
[Section IV] The sentence 'ACORN is purposed as a plug-and-play optimization to exhibit catastrophic failures under distributional shifts' appears to state the opposite of the intended meaning; it should read 'prevent' rather than 'exhibit'.
[Eq. (8)] The notation lambda_KL L_KLD is introduced without definition; Eq. (1) defines lambda and D_KL. Please align the notation and define every symbol in Eq. (8).
[Section IV-A, TDL] TDL is introduced as one of the four safety metrics but is never reported numerically in the experiments; please specify how confidence regions and violations are computed and provide quantitative results.
[Section V-B, Figs. 3-4] The claims about 'more concentrated region of trajectory density' and 'substantial reduction in trajectories violating safety thresholds' are not supported by any quantitative trajectory-density or violation counts; please provide numerical measures.

Circularity Check

2 steps flagged · score 6.0 of 10

ACORN's robustness and safety gains are evaluated using the same Gaussian perturbation family and demonstration-distance metrics that its training objective optimizes, so the headline improvements are partially by construction; additionally, ACR-F depends on an undefined reward, making the 23.1% claim unverifiable.

fitted input called prediction [Eqs. (6)-(7), Secs. IV-B and IV-C; Table I]
"The perturbed action is then defined as: a_noised = (1+β)a_gt, if activated (prob. p); a_gt, otherwise ... Three noise levels are systematically evaluated: LIGHT (p=0.1, σ=0.04), NORMAL (p=0.2, σ=0.06), HEAVY (p=0.3, σ=0.08). ... Negative samples are synthesized through correlated scaling and independent infinitesimal distortions: a− = (1+ησ)⊙a_gt + ϵ ... η∼N(0,1), σ ∈ [0.04, 0.06, 0.08], ϵ∼N(0,δ²I), δ=10⁻⁵."

The evaluation noise in Eq. (6) and the negative-sample generation in Eq. (7) use the same multiplicative Gaussian action-perturbation family with the same σ grid; the only differences are a probabilistic trigger and an infinitesimal additive term. ACORN's training signal explicitly distinguishes clean expert actions from such noised actions, so the Table I robustness gains (including the 23.1% ACR-F improvement under Normal noise) are measured on the perturbation distribution the model was trained to reject. The claim of robustness to 'unpredicted situations and distribution shifts' is therefore partly a check of the training objective rather than an independent generalization result.
self definitional [Eqs. (4)-(5) and Eq. (10), Secs. IV-A.1 and IV-D.2; Table I]
"AM-J = (1/|F|) Σ_{τi∈F} (1/T_i) Σ_{t=0}^{T_i} ||a_t^(i) − a_ref,t^(i)||₂, where a_ref,t = arg min_{a′_t∈D_train} ||a_t^(i) − a′_t||₂. ... LContrast = max(0, ||a+ − â||₂ − ||a− − â||₂ + α)."

AM-J and AM-E quantify the policy's action and end-effector distance to the nearest training demonstration, while the ACT-ACORN objective directly optimizes a closely related quantity: the Huber loss regresses predictions to ground-truth actions and the contrastive term pulls predictions toward the expert positive sample a+. Thus lower AM-J/AM-E for ACORN substantially restates the training objective, and these metrics are reported as 'safety' only because safety was defined as closeness to the training demonstrations. This is not fully forced, since ACT also fits demonstrations, but the safety improvement measured this way is partly self-definitional.

full rationale

The paper contains no self-citation chain or imported uniqueness theorem, and SR is an independent metric: ACT-ACORN matches ACT's SR under Normal noise (62.2 vs 62.2), so the claim of not sacrificing performance has some external content. However, the safety-robustness comparison is entangled with training inputs in two ways. First, Eq. (6) evaluates with multiplicative Gaussian action noise whose σ values are exactly the same as the σ grid used to synthesize negative samples in Eq. (7); the reported robustness gains are therefore in-distribution with respect to ACORN's contrastive training signal. Second, the safety metrics AM-J and AM-E are defined as distance to the nearest training demonstration, and Eq. (10) explicitly optimizes closeness to expert action samples, so reductions in these metrics partly restate the objective. These are partial rather than total circularities because both ACT and ACORN fit demonstrations and because SR is external. Separately, ACR and ACR-F in Eq. (3) require a reward function r_t that is never defined for the LeRobot AlohaTransferCube imitation-learning benchmark, so the headline 23.1% ACR-F improvement cannot be computed or checked; that is a missing-definition and correctness problem rather than a circularity, but it reinforces the assessment that the central quantitative claim is not independently established. The score of 6 reflects that the paper's central safety claims reduce at least partially to the training/evaluation construction, while the absence of self-citation and the presence of an independent SR comparison prevent a higher score.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central evidence rests on hyperparameters tuned on the test condition, an undefined reward, and safety metrics that measure distance to the training distribution. No new physical entities are introduced.

free parameters (5)

Huber threshold delta = 0.124
Set from 'mean loss plus one standard deviation' of pretraining loss, so it is fitted to training statistics.
Contrastive margin alpha = 0.01
Selected by ablation on the Normal noise test condition (Table III).
Curriculum coefficient k = 15
Selected by ablation on the Normal noise test condition (Table II).
Noise activation probability p and intensity sigma = Light p=0.1, sigma=0.04; Normal p=0.2, sigma=0.06; Heavy p=0.3, sigma=0.08
Hand-chosen evaluation and training perturbation levels that define the measured robustness gains.
Independent micro-perturbation scale delta = 1e-5
Chosen by hand; the epsilon term in Eq 7 is negligible relative to the correlated scaling.

assumptions (4)

domain assumption A reward function r_t exists and is computable for failed episodes in the ACT/LeRobot imitation learning setup (used in Eq 3).
ACR-F is the paper's headline metric but r_t is never defined; in pure behavior cloning no reward is used.
domain assumption Closeness to training demonstrations is a valid proxy for safety (used in AM-J and AM-E, Eqs 4-5).
The metrics reward matching the demonstration set, which is also the training target; no independent safety criterion is given.
domain assumption Multiplicative Gaussian noise on ground-truth actions at evaluation simulates the environmental distribution shifts that matter for deployment (Eq 6).
Real distribution shifts also affect observations, dynamics, and contact, none of which are tested.
ad hoc to paper The biomechanical joint taxonomy (high-priority vs secondary joints) determines which joints matter for safety (Section V-B).
The taxonomy is introduced after the results and not supported by controlled experiments.

how reviews work

0 comments

Cite this review

Pith. "Pith review of ACORN: Adaptive Contrastive Optimization for Safe and Robust Fine-Grained Robotic Manipulation." pith.science (2026). https://pith.science/paper/UVGT572J

@misc{pith2026250506628,
  author       = {Pith},
  title        = {Pith review of: ACORN: Adaptive Contrastive Optimization for Safe and Robust Fine-Grained Robotic Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UVGT572J}},
  note         = {Machine review of arXiv:2505.06628}
}

read the original abstract

Embodied AI research has traditionally emphasized performance metrics such as success rate and cumulative reward, overlooking critical robustness and safety considerations that emerge during real-world deployment. In actual environments, agents continuously encounter unpredicted situations and distribution shifts, causing seemingly reliable policies to experience catastrophic failures, particularly in manipulation tasks. To address this gap, we introduce four novel safety-centric metrics that quantify an agent's resilience to environmental perturbations. Building on these metrics, we present Adaptive Contrastive Optimization for Robust Manipulation (ACORN), a plug-and-play algorithm that enhances policy robustness without sacrificing performance. ACORN leverages contrastive learning to simultaneously align trajectories with expert demonstrations while diverging from potentially unsafe behaviors. Our approach efficiently generates informative negative samples through structured Gaussian noise injection, employing a double perturbation technique that maintains sample diversity while minimizing computational overhead. Comprehensive experiments across diverse manipulation environments validate ACORN's effectiveness, yielding improvements of up to 23% in safety metrics under disturbance compared to baseline methods. These findings underscore ACORN's significant potential for enabling reliable deployment of embodied agents in safety-critical real-world applications.

Figures

Figures reproduced from arXiv: 2505.06628 by the authors.

**Figure 2.** The contrastive loss contribution increases exponentially as baseline [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** High-priority Joint Group: (Left) ACT vs. (Right) ACT-ACORN [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Secondary Joint Group: (Left) ACT vs. (Right) ACT-ACORN [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 16 canonical work pages

[1]

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

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

arXiv 2023
[2]

Generative adversarial imitation learning,

J. Ho and S. Ermon, “Generative adversarial imitation learning,” in Advances in Neural Information Processing Systems(NeurIPS) , vol. 29. Curran Associates, Inc., 2016

work page 2016
[3]

Reinforcement learning with augmented data,

M. Laskin, K. Lee, A. Stooke, L. Pinto, P. Abbeel, and A. Srinivas, “Reinforcement learning with augmented data,” Advances in neural information processing systems(NeurIPS), vol. 33, pp. 19 884–19 895, 2020

work page 2020
[4]

A simple frame- work for contrastive learning of visual representations,

T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple frame- work for contrastive learning of visual representations,” in Interna- tional conference on machine learning. PmLR, 2020, pp. 1597–1607

work page 2020
[5]

Catch the ball: Accurate high-speed motions for mobile manipulators via inverse dynamics learning,

K. Dong, K. Pereida, F. Shkurti, and A. P. Schoellig, “Catch the ball: Accurate high-speed motions for mobile manipulators via inverse dynamics learning,” in 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2020, pp. 6718–6725

work page 2020
[6]

Robot reinforcement learning on the constraint manifold,

P. Liu, D. Tateo, H. B. Ammar, and J. Peters, “Robot reinforcement learning on the constraint manifold,” inConference on Robot Learning. PMLR, 2022, pp. 1357–1366

work page 2022
[7]

Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods for robotics and control,

C. Dawson, S. Gao, and C. Fan, “Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods for robotics and control,” IEEE Transactions on Robotics , vol. 39, no. 3, pp. 1749–1767, 2023

2023
[8]

Reinforcement learning for safety-critical control under model uncertainty, using con- trol lyapunov functions and control barrier functions,

J. Choi, F. Castaneda, C. J. Tomlin, and K. Sreenath, “Reinforcement learning for safety-critical control under model uncertainty, using con- trol lyapunov functions and control barrier functions,” arXiv preprint arXiv:2004.07584, 2020

arXiv 2004

Show all 29 references

[9]

Context- aware safe reinforcement learning for non-stationary environments,

B. Chen, Z. Liu, J. Zhu, M. Xu, W. Ding, L. Li, and D. Zhao, “Context- aware safe reinforcement learning for non-stationary environments,” in 2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 10 689–10 695

2021
[10]

Constrained policy optimization via bayesian world models,

Y . As, I. Usmanova, S. Curi, and A. Krause, “Constrained policy optimization via bayesian world models,” arXiv preprint arXiv:2201.09802, 2022

2022 arXiv
[11]

Grasping with chopsticks: Combating covariate shift in model-free imitation learning for fine manipulation,

L. Ke, J. Wang, T. Bhattacharjee, B. Boots, and S. Srinivasa, “Grasping with chopsticks: Combating covariate shift in model-free imitation learning for fine manipulation,” in 2021 IEEE international conference on robotics and automation (ICRA) . IEEE, 2021, pp. 6185–6191

2021
[12]

Ccil: Continuity-based data augmentation for corrective imitation learning,

L. Ke, Y . Zhang, A. Deshpande, S. Srinivasa, and A. Gupta, “Ccil: Continuity-based data augmentation for corrective imitation learning,” arXiv preprint arXiv:2310.12972 , 2023

2023 arXiv
[13]

Data efficient behavior cloning for fine manipulation via continuity- based corrective labels,

A. Deshpande, L. Ke, Q. Pfeifer, A. Gupta, and S. S. Srinivasa, “Data efficient behavior cloning for fine manipulation via continuity- based corrective labels,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2024, pp. 8531– 8538

2024
[14]

An algorithmic perspective on imitation learning,

T. Osa, J. Pajarinen, G. Neumann, J. A. Bagnell, P. Abbeel, J. Peters, et al., “An algorithmic perspective on imitation learning,” Foundations and Trends® in Robotics , vol. 7, no. 1-2, pp. 1–179, 2018

2018
[15]

Algorithms for inverse reinforcement learning

A. Y . Ng, S. Russell, et al. , “Algorithms for inverse reinforcement learning.” in Icml, vol. 1, no. 2, 2000, p. 2

2000
[16]

Learning an approximate model predictive controller with guarantees,

M. Hertneck, J. K ¨ohler, S. Trimpe, and F. Allg ¨ower, “Learning an approximate model predictive controller with guarantees,” IEEE Control Systems Letters , vol. 2, no. 3, pp. 543–548, 2018

2018
[17]

Imitation learning with stability and safety guarantees,

H. Yin, P. Seiler, M. Jin, and M. Arcak, “Imitation learning with stability and safety guarantees,” IEEE Control Systems Letters , vol. 6, pp. 409–414, 2021

2021
[18]

A reduction of imitation learning and structured prediction to no-regret online learning,

S. Ross, G. Gordon, and D. Bagnell, “A reduction of imitation learning and structured prediction to no-regret online learning,” in Proceedings of the fourteenth international conference on artificial intelligence and statistics. JMLR Workshop and Conference Proceedings, 2011, ...

2011
[19]

Efficient reductions for imitation learning,

S. Ross and D. Bagnell, “Efficient reductions for imitation learning,” in Proceedings of the thirteenth international conference on artificial intelligence and statistics . JMLR Workshop and Conference Pro- ceedings, 2010, pp. 661–668

2010
[20]

Is imitation learning the route to humanoid robots?

S. Schaal, “Is imitation learning the route to humanoid robots?” Trends in cognitive sciences , vol. 3, no. 6, pp. 233–242, 1999

1999
[21]

End-to-end driving via conditional imitation learning,

F. Codevilla, M. M ¨uller, A. L ´opez, V . Koltun, and A. Dosovitskiy, “End-to-end driving via conditional imitation learning,” in 2018 IEEE international conference on robotics and automation (ICRA) . IEEE, 2018, pp. 4693–4700

2018
[22]

Models of temporal discounting 1937–2000: An in- terdisciplinary exchange between economics and psychology,

T. Gr ¨une-Yanoff, “Models of temporal discounting 1937–2000: An in- terdisciplinary exchange between economics and psychology,” Science in context, vol. 28, no. 4, pp. 675–713, 2015

1937
[23]

Reinforcement learning with human teachers: Understanding how people want to teach robots,

A. L. Thomaz, G. Hoffman, and C. Breazeal, “Reinforcement learning with human teachers: Understanding how people want to teach robots,” in ROMAN 2006-The 15th IEEE International Symposium on Robot and Human Interactive Communication . IEEE, 2006, pp. 352–357

2006
[24]

Teachable robots: Understanding human teaching behavior to build more effective robot learners,

A. L. Thomaz and C. Breazeal, “Teachable robots: Understanding human teaching behavior to build more effective robot learners,” Artificial Intelligence, vol. 172, no. 6-7, pp. 716–737, 2008

2008
[25]

Dealing with multiple experts and non-stationarity in inverse reinforcement learning: an application to real-life problems,

A. Likmeta, A. M. Metelli, G. Ramponi, A. Tirinzoni, M. Giuliani, and M. Restelli, “Dealing with multiple experts and non-stationarity in inverse reinforcement learning: an application to real-life problems,” Machine Learning, vol. 110, pp. 2541–2576, 2021

2021
[26]

One-shot imitation in a non-stationary environment via multi-modal skill,

S. Shin, D. Lee, M. Yoo, W. K. Kim, and H. Woo, “One-shot imitation in a non-stationary environment via multi-modal skill,” inInternational Conference on Machine Learning . PMLR, 2023, pp. 31 562–31 578

2023
[27]

Imitation learning: Progress, taxonomies and challenges,

B. Zheng, S. Verma, J. Zhou, I. W. Tsang, and F. Chen, “Imitation learning: Progress, taxonomies and challenges,” IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 5, pp. 6322–6337, 2022

2022
[28]

Curriculum learning,

Y . Bengio, J. Louradour, R. Collobert, and J. Weston, “Curriculum learning,” in Proceedings of the 26th annual international conference on machine learning , 2009, pp. 41–48

2009
[29]

Lerobot: State-of-the-art machine learning for real-world robotics in pytorch,

R. Cadene, S. Alibert, A. Soare, Q. Gallouedec, A. Zouitine, and T. Wolf, “Lerobot: State-of-the-art machine learning for real-world robotics in pytorch,” https://github.com/huggingface/lerobot, 2024

2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

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

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

arXiv 2023

[2] [2]

Generative adversarial imitation learning,

J. Ho and S. Ermon, “Generative adversarial imitation learning,” in Advances in Neural Information Processing Systems(NeurIPS) , vol. 29. Curran Associates, Inc., 2016

work page 2016

[3] [3]

Reinforcement learning with augmented data,

M. Laskin, K. Lee, A. Stooke, L. Pinto, P. Abbeel, and A. Srinivas, “Reinforcement learning with augmented data,” Advances in neural information processing systems(NeurIPS), vol. 33, pp. 19 884–19 895, 2020

work page 2020

[4] [4]

A simple frame- work for contrastive learning of visual representations,

T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple frame- work for contrastive learning of visual representations,” in Interna- tional conference on machine learning. PmLR, 2020, pp. 1597–1607

work page 2020

[5] [5]

Catch the ball: Accurate high-speed motions for mobile manipulators via inverse dynamics learning,

K. Dong, K. Pereida, F. Shkurti, and A. P. Schoellig, “Catch the ball: Accurate high-speed motions for mobile manipulators via inverse dynamics learning,” in 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2020, pp. 6718–6725

work page 2020

[6] [6]

Robot reinforcement learning on the constraint manifold,

P. Liu, D. Tateo, H. B. Ammar, and J. Peters, “Robot reinforcement learning on the constraint manifold,” inConference on Robot Learning. PMLR, 2022, pp. 1357–1366

work page 2022

[7] [7]

Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods for robotics and control,

C. Dawson, S. Gao, and C. Fan, “Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods for robotics and control,” IEEE Transactions on Robotics , vol. 39, no. 3, pp. 1749–1767, 2023

2023

[8] [8]

Reinforcement learning for safety-critical control under model uncertainty, using con- trol lyapunov functions and control barrier functions,

J. Choi, F. Castaneda, C. J. Tomlin, and K. Sreenath, “Reinforcement learning for safety-critical control under model uncertainty, using con- trol lyapunov functions and control barrier functions,” arXiv preprint arXiv:2004.07584, 2020

arXiv 2004