Pith. sign in

REVIEW 4 major objections 5 minor 35 references

Should We Learn Contact-Rich Manipulation Policies from Sampling-Based Planners?

T0 review · 4 major / 5 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 The paper claims that behavior cloning for contact-rich manipulation can succeed with model-based planner demonstrations only when those demonstrations are consistent and low-entropy, and that greedy and PRM planners designed for…

desk verdict A useful empirical study on planner-generated demos for BC, but the headline causal claim about entropy is not actually isolated. read the letter →

arxiv 2412.09743 v3 pith:WCPZ6RER submitted 2024-12-12 cs.RO

Huaijiang Zhu , Tong Zhao , Xinpei Ni , Jiuguang Wang , Kuan Fang , Ludovic Righetti , Tao Pang This is my paper · ORCID
classification cs.RO
keywords behaviorcloningcontact-richmanipulationsampling-basedplanningactionentropydemonstrationconsistencydiffusionpolicygoal-conditionedimitationlearningdexterous
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

This paper asks whether sampling-based motion planners can replace human teleoperation as the source of training data for contact-rich manipulation policies. It argues that they can, but only when the planner produces consistent, low-entropy demonstrations: standard RRT planners generate highly variable action distributions that degrade behavior cloning, especially with few demonstrations. The paper shows that swapping RRT for a greedy planner on a bimanual rotation task, and for a fixed-primitive PRM planner on an in-hand reorientation task, yields policies with near-perfect simulation success and successful zero-shot transfer to hardware. The central claim is that demonstration consistency matters more than state coverage when learning from planned data.

What carries the argument

The load-bearing machinery is the data-generation pipeline built around a quasi-dynamic contact planner with smoothed contact dynamics, where each step solves a convex optimization problem (Problem 1) to move the object toward a desired pose. The paper modifies this planner twice to lower demonstration entropy: a greedy search that repeatedly solves the same optimization without stochastic subgoal sampling, and a sparse probabilistic roadmap whose nodes are 24 canonical cube orientations connected by three fixed primitives (PitchPlus90, YawPlus45, YawMinus45). These plans are rolled out in a physics simulator without contact smoothing and labeled with hindsight goal relabeling, and a goal-conditioned diffusion policy (a 1D U-Net with FiLM conditioning) is trained to imitate the resulting state-action chunks. Action entropy is the organizing quantity throughout: the paper measures it as entropy of discretized velocity directions, spread of per-segment goal progress, and regrasp-timing entropy.

What would settle it

Train the same behavior-cloning policy on a low-entropy planner's demonstrations with calibrated random action noise added to match RRT's measured entropy, holding state coverage and path length fixed; if success does not drop as entropy increases, the paper's causal attribution is wrong.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the entropy of planner-generated demonstrations, not their state-space coverage, determines whether a behavior-cloned policy can learn a contact-rich skill. The authors characterize action entropy through three proxies—discretized object velocity direction, per-contact-segment progress toward the goal, and regrasp timing—and show that RRT-based contact planning produces consistently higher entropy on all three. Policies trained on lower-entropy data from a greedy planner reach 99–100% success on the bimanual IiwaBimanual task with 100–5000 demonstrations, while RRT-trained policies plateau around 84–88% success. For the 16-DoF AllegroHand in-hand reorientation task, a PRM planner built from 24 canonical cube orientations and three fixed rotation primitives supports 68–82% simulation success and 62.5% hardware success, with the bimanual policy reaching 90% hardware success. The paper concludes that prioritizing demonstration consistency while maintaining solution diversity is the key design principle for using model-based planners as behavior-cloning teachers.

Load-bearing premise

The load-bearing premise is that the measured performance differences between planners are caused by action-entropy differences, not by other correlated differences such as path optimality, state-distribution narrowness, or task-specific planner engineering.

Editorial extensions

If this is right

  • Replacing human teleoperation with model-based planners is viable for contact-rich manipulation if the planner is designed for demonstration consistency.
  • Low action entropy in demonstrations, not broad state coverage, is the main driver of behavior-cloning success in the low-data regime.
  • Task-specific planner structure—greedy search for planar tasks and a PRM with fixed primitives for dexterous in-hand tasks—can make planned data learnable.
  • A diffusion-based goal-conditioned policy trained on such data can transfer zero-shot to hardware, achieving 90% success on the bimanual task and 62.5% on in-hand reorientation.

Reading between the lines

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

  • A testable extension is to filter RRT demonstrations by an entropy threshold: if success improves with filtering, entropy is the causal variable; if not, other planner differences matter.
  • The consistency-over-coverage principle likely extends beyond planners to any demonstration source, including teleoperation and reinforcement learning, where per-state action variance should be measured and controlled.
  • The PRM-plus-adjustment-policy recipe suggests a modular design: hard constraints encoded in a low-entropy planner, with a small learned policy handling the variable final approach to the goal.
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.

Referee Report

4 major / 5 minor

Summary. The paper investigates whether sampling-based motion planners can serve as data generators for behavior cloning (BC) of contact-rich manipulation skills. It proposes a pipeline that combines a contact-rich planner (RRT, greedy search, or a PRM variant) with simulation rollouts, hindsight goal relabeling, and a diffusion-based goal-conditioned policy. The central empirical claim is that RRT-generated demonstrations have high action entropy, measured through proxy statistics, and that this degrades BC performance, whereas more consistent demonstrations from greedy search (IiwaBimanual) or a PRM-based planner (AllegroHand) yield substantially better policies, including zero-shot hardware transfer. The paper reports success rates for multiple dataset sizes, entropy visualizations, and hardware evaluations for both tasks.

Significance. If the central claim held, the paper would provide actionable guidance for scaling BC beyond teleoperation: instead of maximizing demonstration coverage, planners should be designed to produce low-entropy, consistent demonstrations. The paper's strengths are its concrete and reproducible pipeline, the use of two challenging contact-rich tasks, the dataset-size sweep in Table I, and genuine zero-shot hardware transfer for both tasks. The planner algorithms are described in sufficient detail to reimplement. However, the causal attribution to action entropy is not established by the current experiments, and several load-bearing comparisons lack uncertainty quantification. The pipeline itself is a useful empirical contribution even if the entropy hypothesis is later refined.

major comments (4)
  1. [Section III-A2, Table I] The headline comparison in Table I reports a single success-rate number per planner and dataset size, with no repeated seeds, error bars, or significance tests. Diffusion policy training is stochastic, and the evaluation uses 100 random initial object poses, so the gap between 44% and 99% at 100 demonstrations, and the claimed RRT plateau near 85%, have no stated uncertainty. Please report results over at least three training seeds with mean and standard deviation, or otherwise quantify the variability of the comparison.
  2. [Section III-A2] The entropy measures are not estimates of the conditional action distribution p(A|O,g) that the BC policy actually learns. The paper computes Shannon entropy of discretized object velocity directions, per-segment progress, and regrasp Bernoulli events; these are aggregate statistics. A planner could have low velocity-direction entropy yet high conditional action variance given the same observation and goal, or vice versa. The causal claim of contribution 1 therefore is not directly tested. I recommend either estimating conditional entropy (e.g., by binning observations and goals and measuring action variance within each bin) or designing an experiment that varies demonstration entropy while holding the planner fixed, such as by resampling or reweighting trajectories from the same planner.
  3. [Section III-A2 and Section III-B] The RRT versus greedy/PRM comparisons vary many attributes simultaneously, not just entropy. RRT trajectories are longer, meander, contain negative-progress segments, and have different regrasp timing and frequency; the AllegroHand PRM pipeline additionally replaces the planner with precomputed primitives and fixed grasps, and the final system uses a hybrid policy with an open-hand reset. Any of these differences could explain the policy performance gap. The paper's own discussion in Section IV-C attributes failures to out-of-distribution scenarios not present in training data, which points toward coverage and planner structure rather than entropy per se. Please isolate the entropy variable or reframe contribution 1 as a correlational observation and present the pipeline itself as the main contribution.
  4. [Section III-A2 and Section IV-C] The test distribution is narrower than the training distribution: the text states that the randomization range is 'slightly shrunken' at test time so that the training set has broader coverage. This protocol can make a consistent planner look better if the two planners have different state coverage, and it limits the generality of the reported success rates. Please report performance on the full training-range distribution as well, or justify the shrunken test set as the intended deployment distribution.
minor comments (5)
  1. [Section III-B.3] There is a typo: 'kept the ssame as' should be 'kept the same as'.
  2. [Section III-A2] The text says 'the Bernouli event' and 'B = 2'; please correct the spelling to 'Bernoulli' and clarify that B denotes the number of bins.
  3. [Section III-B.1] The description of the PRM planner says 'traverse through all other canonical orientations'; please clarify whether the planner visits all 24 canonical orientations or only a subset, since this affects trajectory length and the interpretation of the entropy comparison.
  4. [Section II-B] The diffusion training loss is written as MSE(epsilon_k, epsilon_theta(...)); a brief description of the noise schedule and the number of diffusion steps would help reproducibility, beyond the reference to [6].
  5. [Figure 5] The caption says 'We show 100 trajectories for each task to avoid cluttering,' but the panels appear to show many overlapping curves; please state explicitly which curves correspond to which planner in each panel, since panels (d) and (e) are easy to confuse.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the high-entropy claim is empirical and independently measured; self-cited planners are used as tools, not as the basis for the conclusion.

full rationale

The paper's central claim (contribution 1) is an empirical observation: datasets generated by the RRT planner have higher measured entropy on the paper's own proxies (velocity-direction, segment-progress, and regrasp entropies) and, when used to train the same diffusion BC architecture, yield lower success rates than datasets from the greedy or PRM planners. No parameter is fitted to the success numbers, and success is not defined in terms of entropy; the entropy metrics and the policy evaluations are separate measurements. The planners from [21] and [24] involve overlapping authors, but the paper does not rely on those citations to establish the empirical result: Algorithm 1 is given in the paper, the PRM construction is described, and the entropy/success comparisons are conducted on data generated by the described procedures. The causal attribution of the performance gap to action entropy is open to confounds, since the compared planners differ in trajectory geometry, regrasp structure, and state coverage, but that is an experimental-validity concern rather than a circularity: the paper does not reduce the outcome to its input by construction, and no fitted parameter is renamed as a prediction. Therefore no significant circularity is present.

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

The central claim rests on the quasi-dynamic and smoothed-contact planning assumptions borrowed from [21], [24], and on the entropy proxy measures introduced in this paper. No new physical entities are introduced.

free parameters (4)
  • pgrasp = not reported
    Probability of sampling a new grasp in Contact RRT (Algorithm 1); affects demonstration diversity and is not specified.
  • weight in progress metric = 0.2
    Coefficient for rotation error in D(q,qgoal) used in segment progress analysis; arbitrary.
  • entropy binning parameters = 16 quadrants, 3 angular classes, 25 time bins, 0.05m grid
    Discretization choices in Section III-A2 to estimate action entropy; influence the reported H values.
  • dataset sizes = 100/500/1000/5000 for IB; 1000+5000 for AH
    Chosen dataset scales and the AH train split (1000 global + 5000 adjustment) are hand-selected.
assumptions (4)
  • domain assumption Quasi-dynamic system dynamics (velocity and acceleration negligible)
    Used in planning through contact (Section II-A.1) based on [28]; simplies state to joint positions and object pose.
  • domain assumption Smoothed contact dynamics with convex trust region linearization
    Problem (1) is solved with smoothed contacts and linearization from [21], [24]; enables convex optimization but introduces plan-rollout discrepancies.
  • domain assumption Any SO(3) orientation is within 63° of a canonical orientation (Monte Carlo estimate)
    Used in Section III-B.1 to design the PRM; justifies coverage by the 24 canonical nodes.
  • domain assumption Hindsight goal relabeling yields valid goals from reached states
    Standard GCIL assumption [29]-[31] used in Section II-B; relies on the demonstration implying a valid trajectory to the relabeled goal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Should We Learn Contact-Rich Manipulation Policies from Sampling-Based Planners?." pith.science (2026). https://pith.science/paper/WCPZ6RER

@misc{pith2026241209743,
  author       = {Pith},
  title        = {Pith review of: Should We Learn Contact-Rich Manipulation Policies from Sampling-Based Planners?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WCPZ6RER}},
  note         = {Machine review of arXiv:2412.09743}
}
read the original abstract

The tremendous success of behavior cloning (BC) in robotic manipulation has been largely confined to tasks where demonstrations can be effectively collected through human teleoperation. However, demonstrations for contact-rich manipulation tasks that require complex coordination of multiple contacts are difficult to collect due to the limitations of current teleoperation interfaces. We investigate how to leverage model-based planning and optimization to generate training data for contact-rich dexterous manipulation tasks. Our analysis reveals that popular sampling-based planners like rapidly exploring random tree (RRT), while efficient for motion planning, produce demonstrations with unfavorably high entropy. This motivates modifications to our data generation pipeline that prioritizes demonstration consistency while maintaining solution diversity. Combined with a diffusion-based goal-conditioned BC approach, our method enables effective policy learning and zero-shot transfer to hardware for two challenging contact-rich manipulation tasks.

Figures

Figures reproduced from arXiv: 2412.09743 by the authors.

Figure 1
Figure 1. Contact-rich manipulation tasks considered in this work. Top: AllegroHand: In-hand re-orientation with a dexterous hand. Bottom: IiwaBimanual: Bimanual manipulation of a cylinder. Abstract— The tremendous success of behavior cloning (BC) in robotic manipulation has been largely confined to tasks where demonstrations can be effectively collected through human teleoperation. However, demonstrations for contact-rich ma… view at source ↗
Figure 2
Figure 2. provides an overview of our method. First, we obtain a training dataset via a multi-stage data curation pipeline: 1) a model-based planner using smoothed contact dynamics [21], [25]–[27] proposes a plan, 2) this plan is then executed in a physics simulator without contact smoothing to produce state-action trajectories. 3) the reached states are labeled as goals using hindsight goal relabeling. Using the generated da… view at source ↗
Figure 3
Figure 3. Entropy of linear and angular velocity directions of the RRT and greedy datasets for (a) IiwaBimaual (IB) and (b) AllegroHand (AH). White indicates that there is no data [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Example demonstrations for IiwaBimanual (IB) and Alle￾groHand(AH). In all subfigures, the solid frames indicate the goal object configuration. For both tasks, contact RRT covers more space by following a more meandering path before reaching the goal than their lower-en…
Figure 5
Figure 5. Figure 5: Normalized weighted distance to goal of the object is plotted against trajectory completion percentage for (a) contact RRT IiwaBimanual (IB) and (b) greedy search IB. Each colored curve represents one demonstration trajectory. Dots along a curve represent regrasps. Gra…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 11 canonical work pages

  1. [13]

    Data quality in imitation learn- ing,

    S. Belkhale, Y . Cui, and D. Sadigh, “Data quality in imitation learn- ing,” Advances in Neural Information Processing Systems , vol. 36, 2024

    2024

  2. [1]

    Bc-z: Zero-shot task generalization with robotic imitation learning,

    E. Jang, A. Irpan, M. Khansari, D. Kappler, F. Ebert, C. Lynch, S. Levine, and C. Finn, “Bc-z: Zero-shot task generalization with robotic imitation learning,” in Conference on Robot Learning. PMLR, 2022, pp. 991–1002

    2022

  3. [2]

    Cliport: What and where pathways for robotic manipulation,

    M. Shridhar, L. Manuelli, and D. Fox, “Cliport: What and where pathways for robotic manipulation,” in Conference on robot learning . PMLR, 2022, pp. 894–906

    2022

  4. [3]

    Diffusion policy: Visuomotor policy learning via action diffusion,

    C. Chi, S. Feng, Y . Du, Z. Xu, E. Cousineau, B. Burchfiel, and S. Song, “Diffusion policy: Visuomotor policy learning via action diffusion,” arXiv preprint arXiv:2303.04137 , 2023

    arXiv 2023

  5. [4]

    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

  6. [5]

    Aloha unleashed: A simple recipe for robot dexterity,

    T. Z. Zhao, J. Tompson, D. Driess, P. Florence, K. Ghasemipour, C. Finn, and A. Wahid, “Aloha unleashed: A simple recipe for robot dexterity,” arXiv preprint arXiv:2410.13126 , 2024

    arXiv 2024

  7. [6]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems , vol. 33, pp. 6840–6851, 2020

    2020

  8. [7]

    Goal-conditioned imi- tation learning using score-based diffusion policies,

    M. Reuss, M. Li, X. Jia, and R. Lioutikov, “Goal-conditioned imi- tation learning using score-based diffusion policies,” arXiv preprint arXiv:2304.02532, 2023

    arXiv 2023

Show all 35 references
  1. [8]

    Universal manipulation interface: In-the-wild robot teaching without in-the-wild robots,

    C. Chi, Z. Xu, C. Pan, E. Cousineau, B. Burchfiel, S. Feng, R. Tedrake, and S. Song, “Universal manipulation interface: In-the-wild robot teaching without in-the-wild robots,”arXiv preprint arXiv:2402.10329, 2024

    2024 arXiv

  2. [9]

    Scaling laws for neural language models,

    J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei, “Scaling laws for neural language models,” arXiv preprint arXiv:2001.08361 , 2020

    2001 arXiv

  3. [10]

    End-to- end urban driving by imitating a reinforcement learning coach,

    Z. Zhang, A. Liniger, D. Dai, F. Yu, and L. Van Gool, “End-to- end urban driving by imitating a reinforcement learning coach,” in Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 15 222–15 232

    2021

  4. [11]

    Learning robust perceptive locomotion for quadrupedal robots in the wild,

    T. Miki, J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter, “Learning robust perceptive locomotion for quadrupedal robots in the wild,” Science Robotics, vol. 7, no. 62, p. eabk2822, 2022

    2022

  5. [12]

    Visual dexterity: In-hand reorientation of novel and complex object shapes,

    T. Chen, M. Tippur, S. Wu, V . Kumar, E. Adelson, and P. Agrawal, “Visual dexterity: In-hand reorientation of novel and complex object shapes,” Science Robotics, vol. 8, no. 84, p. eadc9244, 2023

    2023

  6. [14]

    In-hand object rotation via rapid motor adaptation,

    H. Qi, A. Kumar, R. Calandra, Y . Ma, and J. Malik, “In-hand object rotation via rapid motor adaptation,” in Conference on Robot Learning. PMLR, 2023, pp. 1722–1732

    2023

  7. [15]

    Rotating without seeing: Towards in-hand dexterity through touch,

    Z.-H. Yin, B. Huang, Y . Qin, Q. Chen, and X. Wang, “Rotating without seeing: Towards in-hand dexterity through touch,” inRobotics: Science and Systems, 2023

    2023

  8. [16]

    Dextreme: Transfer of agile in-hand manipulation from simu- lation to reality,

    A. Handa, A. Allshire, V . Makoviychuk, A. Petrenko, R. Singh, J. Liu, D. Makoviichuk, K. Van Wyk, A. Zhurkevich, B. Sundaralingam, et al., “Dextreme: Transfer of agile in-hand manipulation from simu- lation to reality,” in 2023 IEEE International Conference on Robotics and Au...

    2023

  9. [17]

    Neural mp: A generalist neural motion planner,

    M. Dalal, J. Yang, R. Mendonca, Y . Khaky, R. Salakhutdinov, and D. Pathak, “Neural mp: A generalist neural motion planner,” arXiv preprint arXiv:2409.05864, 2024

    2024 arXiv

  10. [18]

    Learning locomotion skills from mpc in sensor space,

    M. Khadiv, A. Meduri, H. Zhu, L. Righetti, and B. Sch ¨olkopf, “Learning locomotion skills from mpc in sensor space,” in Learning for Dynamics and Control Conference. PMLR, 2023, pp. 1218–1230

    2023

  11. [19]

    Trajectotree: Trajectory optimization meets tree search for planning multi-contact dexterous manipulation,

    C. Chen, P. Culbertson, M. Lepert, M. Schwager, and J. Bohg, “Trajectotree: Trajectory optimization meets tree search for planning multi-contact dexterous manipulation,” in 2021 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS) . IEEE, 2021, pp. 8262–8268

    2021

  12. [20]

    Enhancing dexterity in robotic manipulation via hierarchical contact exploration,

    X. Cheng, S. Patil, Z. Temel, O. Kroemer, and M. T. Mason, “Enhancing dexterity in robotic manipulation via hierarchical contact exploration,” IEEE Robotics and Automation Letters , vol. 9, no. 1, pp. 390–397, 2023

    2023

  13. [21]

    Global planning for contact-rich manipulation via local smoothing of quasi-dynamic contact models,

    T. Pang, H. T. Suh, L. Yang, and R. Tedrake, “Global planning for contact-rich manipulation via local smoothing of quasi-dynamic contact models,” IEEE Transactions on Robotics , 2023

    2023

  14. [22]

    Efficient object manipulation planning with monte carlo tree search,

    H. Zhu, A. Meduri, and L. Righetti, “Efficient object manipulation planning with monte carlo tree search,” in 2023 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS) . IEEE, 2023, pp. 10 628–10 635

    2023

  15. [23]

    Towards tight convex relaxations for contact-rich manipulation,

    B. P. Graesdal, S. Y . C. Chia, T. Marcucci, S. Morozov, A. Amice, P. A. Parrilo, and R. Tedrake, “Towards tight convex relaxations for contact-rich manipulation,” arXiv preprint arXiv:2402.10312 , 2024

    2024 arXiv

  16. [24]

    Dexterous contact-rich manipulation via the contact trust region,

    H. T. Suh, T. Pang, T. Zhao, and R. Tedrake, “Dexterous contact-rich manipulation via the contact trust region,” 2025. [Online]. Available: https://ctr.theaiinstitute.com/

    2025

  17. [25]

    Dojo: A differentiable physics engine for robotics,

    T. A. Howell, S. L. Cleac’h, J. Br ¨udigam, J. Z. Kolter, M. Schwa- ger, and Z. Manchester, “Dojo: A differentiable physics engine for robotics,” arXiv preprint arXiv:2203.00806 , 2022

    2022 arXiv

  18. [26]

    Bundled gradients through contact via randomized smoothing,

    H. J. T. Suh, T. Pang, and R. Tedrake, “Bundled gradients through contact via randomized smoothing,” IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 4000–4007, 2022

    2022

  19. [27]

    Do dif- ferentiable simulators give better policy gradients?

    H. J. Suh, M. Simchowitz, K. Zhang, and R. Tedrake, “Do dif- ferentiable simulators give better policy gradients?” in International Conference on Machine Learning . PMLR, 2022, pp. 20 668–20 696

    2022

  20. [28]

    A convex quasistatic time-stepping scheme for rigid multibody systems with contact and friction,

    T. Pang and R. Tedrake, “A convex quasistatic time-stepping scheme for rigid multibody systems with contact and friction,” in 2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 6614–6620

    2021

  21. [29]

    Goal-conditioned imitation learning,

    Y . Ding, C. Florensa, P. Abbeel, and M. Phielipp, “Goal-conditioned imitation learning,” Advances in neural information processing sys- tems, vol. 32, 2019

    2019

  22. [30]

    Relay policy learning: Solving long-horizon tasks via imitation and reinforce- ment learning,

    A. Gupta, V . Kumar, C. Lynch, S. Levine, and K. Hausman, “Relay policy learning: Solving long-horizon tasks via imitation and reinforce- ment learning,” in Conference on Robot Learning , 2020

    2020

  23. [31]

    Learning to reach goals via iterated supervised learning,

    D. Ghosh, A. Gupta, A. Reddy, J. Fu, C. Devin, B. Eysenbach, and S. Levine, “Learning to reach goals via iterated supervised learning,” arXiv preprint arXiv:1912.06088 , 2019

    1912 arXiv

  24. [32]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, pro- ceedings, part III 18 ...

    2015

  25. [33]

    Film: Visual reasoning with a general conditioning layer,

    E. Perez, F. Strub, H. De Vries, V . Dumoulin, and A. Courville, “Film: Visual reasoning with a general conditioning layer,” in Proceedings of the AAAI conference on artificial intelligence , vol. 32, no. 1, 2018

    2018

  26. [34]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” arXiv preprint arXiv:1711.05101 , 2017

    2017 arXiv

  27. [35]

    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

Pith tools

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