Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

SPECI: Skill Prompts based Hierarchical Continual Imitation Learning for Robot Manipulation

T0 review · 5 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A hierarchical policy that stores skills as prompts in an expandable codebook outperforms existing continual imitation learning on all four LIBERO task suites.

desk verdict SPECI is a plausible architecture for continual robot learning, but the main table contains an impossible FWT number and a key interference-avoidance claim is undocumented. read the letter →

arxiv 2504.15561 v1 pith:XGLWY2HG submitted 2025-04-22 cs.RO cs.LG

Jingkai Xu , Xiangli Nie This is my paper · ORCID
classification cs.ROcs.LG
keywords continualimitationlearningrobotmanipulationskillpromptshierarchicalpolicyexpandablecodebookmodeapproximationbidirectionalknowledgetransferLIBERO
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 proposes SPECI, an end-to-end hierarchical continual imitation learning policy for robot manipulation, and claims it outperforms existing continual imitation learning methods across all metrics on four LIBERO task suites. The design treats skills as learnable prompts in an expandable codebook: each new task adds skill vectors instead of overwriting old ones, and each step composes a latent skill by attention-weighted selection from the codebook. A 'mode approximation' mechanism factorizes transformer attention weights into shared and task-specific parts, so tasks can share declarative knowledge while keeping their own parameters. If these claims hold, lifelong robot learning needs no manual skill definitions or full retraining, and new tasks can even sharpen old ones.

What carries the argument

The central machinery is an expandable skill codebook plus a mode-approximation mechanism. Each task contributes M skill vectors (with keys and attention vectors) to the codebook; old subsets are frozen, and at each timestep the policy uses the state embedding as a query, computes cosine similarity against all keys, takes the top-C vectors, and combines them via softmax weights into a synthesized latent skill $\tilde{p}$ that is split into key/value prefixes injected into transformer cross-attention. The mode approximation factorizes the transformer's attention weight tensor into shared factors $U,V$ and task-specific factors $Q_k,\lambda_k$ via CP decomposition, adding a low-rank task-specific perturbation to the attention weights. Together these carry skill-level and task-level transfer.

What would settle it

Train SPECI on a LIBERO task sequence with the Schmidt orthogonalization step disabled while keeping all other settings fixed; if the reported negative NBT values degrade substantially, the interference-prevention assumption is load-bearing, and if they do not, the assumption is not doing the work claimed.

Watch

Extended reading notes

Core claim

The paper's central claim is that a hierarchical policy which treats skills as learnable prompts stored in an expandable codebook, and which augments transformer attention with shared plus task-specific parameters, can continually acquire new manipulation tasks while preserving—and even improving—performance on earlier tasks. On the four LIBERO suites, the authors report that SPECI outperforms all compared continual imitation learning baselines on forward transfer (FWT), negative backward transfer (NBT), and overall area under the success curve (AUC), and under the parameter-isolation paradigm it achieves negative NBT, meaning later tasks improve earlier-task success. The authors interpret this as evidence of bidirectional knowledge transfer at both skill and task levels.

Load-bearing premise

The load-bearing premise is that freezing old skill vectors and applying Schmidt orthogonalization prevents old and new skills from interfering in the shared codebook; the paper asserts this without proof or ablation.

Editorial extensions

If this is right

  • A robot policy can be trained sequentially on new manipulation tasks without erasing prior skills, because old skill vectors stay frozen and are reused through attention selection.
  • Later tasks can improve earlier ones: negative NBT under the parameter-isolation paradigm implies backward transfer, so the model keeps refining old behavior as new skills are added.
  • Skill selection is composed from multiple top-C vectors rather than a single primitive, which lets the policy blend skills for long-horizon tasks that combine several sub-actions.
  • Task-level knowledge is isolated by mode approximation, so tasks with different goals need not overwrite shared task parameters.
  • The same unified policy handles perception, skill inference, and action generation end-to-end, avoiding a separate planner or manual skill library.

Reading between the lines

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

  • An untested implication is that the codebook alone might provide rehearsal-free retention, since the paper notes its current implementation relies on replay-based or parameter-isolation training paradigms; disabling those and measuring NBT would isolate the codebook's contribution.
  • The top-C attention weights could be probed to see whether skill subsets correspond to interpretable sub-behaviors such as grasp or push; the visualizations suggest reuse patterns but stop short of quantifying this correspondence.
  • The paper asserts that Schmidt orthogonalization prevents key-skill interference across tasks but does not report applying it; if the step is omitted, the observed negative NBT would be attributable to freezing old vectors and the task-specific mode factors rather than the stated constraint.
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.

Referee Report

5 major / 6 minor

Summary. The paper proposes SPECI, an end-to-end hierarchical continual imitation learning (CIL) framework for robot manipulation. SPECI combines a multimodal perception and fusion module (CLIP language, ResNet-18 vision with FiLM, MLP proprioception) with a high-level skill inference module that maintains an expandable skill codebook, selects top-C skills via an attention-driven cosine-similarity mechanism, and injects the synthesized latent skill into a temporal transformer via prefix tuning. A low-level action execution module uses a GMM policy head. The paper further introduces 'mode approximation,' a CP-decomposition-based parameter-efficient adaptation that augments attention weights with task-specific and task-sharing factors. SPECI is evaluated on the four LIBERO task suites under two lifelong learning paradigms (ER and PACKNET), reporting FWT, NBT, and AUC, with comparisons to flat baselines, BUDS, LOTUS, SEQUENTIAL, and MULTITASK upper bounds. The central claim is that SPECI achieves state-of-the-art bidirectional knowledge transfer and superior overall performance.

Significance. If the reported results hold, the paper makes a useful empirical contribution: it demonstrates that a hierarchical policy with an online-expanding skill codebook and mode-approximated attention can improve both forward and backward transfer on a standard robot manipulation CIL benchmark. The evaluation follows the established LIBERO protocol with three seeds, and the reported gains over LOTUS and BUDS are large and mostly consistent. The component ablation in Table III provides some evidence for the contribution of the codebook and mode approximation. However, the manuscript contains internal inconsistencies in the definition and use of the SEQUENTIAL upper bound, incomplete baseline reporting in Table I, and a central algorithmic claim (Schmidt orthogonalization) that is stated but neither specified nor ablated. These issues must be resolved before the experimental claims can be considered reliable.

major comments (5)
  1. [Section IV-B, Table I, Eq. (12), Section IV-A.3] The reported FWT for SPECI under ER on LIBERO-OBJECT (0.83 ± 0.01) and LIBERO-LONG (0.58 ± 0.01) exceeds the corresponding SEQUENTIAL baseline values (0.80 ± 0.02 and 0.57 ± 0.00). Section IV-A.3 explicitly defines SEQUENTIAL as 'the upper bound for FWT as it fully adapts to each new task,' and Fig. 5 uses SEQUENTIAL as the upper bound. This is internally inconsistent: no continual method with replay should have a higher area under the success-rate curve on new tasks than a method that fully fine-tunes on each new task without replay. The authors must reconcile this, either by correcting the definition or protocol of the SEQUENTIAL baseline (e.g., early-stopping behavior), or by re-computing the FWT numbers. The text in Section IV-B that describes a '4% performance surplus beyond the upper bound on LIBERO-LONG' directly depends on this unresolved contradiction.
  2. [Section III-C, Eqs. (4)-(6)] The skill selection mechanism is not reproducible as written because of a dimensional mismatch. The state embedding is defined as s_e^t ∈ R^{B×L×d} while the attention vectors A and keys K are defined as R^{B×m×d}. Equation (5) computes the Hadamard product s_e^t ⊙ A, which is undefined when L ≠ m (in this paper L=50 and m=10 for the first task). The authors need to specify how the temporal state dimension is reduced or broadcast to align with the m skill keys. In addition, the claim that 'applying Schmidt orthogonalization before learning new tasks is enough to prevent interference between key and skill vectors acquired across different tasks' is asserted in the text with the phrase 'we found' but no algorithm, no implementation detail, and no ablation is provided. This is load-bearing because the top-C selection in Eq. (6) determines how old and new skills are mixed; without evidence that orthogonalization is actually applied and effective, the reported bidirectional transfer could be contaminated by key/skill collisions.
  3. [Table I, LIBERO-SPATIAL and LIBERO-LONG blocks] Table I reports BUDS and LOTUS results only for LIBERO-OBJECT and LIBERO-GOAL; the LIBERO-SPATIAL and LIBERO-LONG blocks contain only the flat baselines, SEQUENTIAL, and SPECI. Yet the text in Section IV-B claims comparisons 'across all four task suites' and reports, for example, a '21% reduction in NBT on LIBERO-SPATIAL compared to the best previous result.' If BUDS and LOTUS were evaluated on these suites, the results must be shown; if not, the text and the abstract's claim of outperforming state-of-the-art CIL methods across all evaluated metrics are unsupported for two of the four suites. This is a reporting omission that affects the main claim.
  4. [Section IV-C, Table III] In the ablation study, the row 'ResNet-T w. Hier. simply duplicates the transformer decoder.' This control increases model capacity and adds a hierarchical decomposition in name only; it does not include the skill inference module or the latent skill variable z_t that defines SPECI's hierarchy in Eq. (3). Consequently, the comparison between 'ResNet-T w. Hier.' and 'SPECI' cannot isolate the effect of hierarchical skill inference; it conflates architectural capacity with the proposed skill-selection mechanism. An ablation that removes the skill codebook but retains a latent-variable hierarchical policy (e.g., a randomly initialized latent variable without codebook selection) is needed to support the claim that the hierarchical architecture itself contributes to the gains.
  5. [Section III-E, Eqs. (10)-(11)] The mode approximation component is not well defined. Equation (10) factorizes W^k ∈ R^{d×d×N} into a sum of rank-one components u_r ∘ v_r ∘ q_r. Equation (11) then writes H^k = W_o X^k + (Σ_r λ^k_r (u_r ∘ v_r ∘ q_r)) X^k. It is unclear how the 3D tensor (u_r ∘ v_r ∘ q_r) contracts with X^k, which is described as an input tensor (either s_t^{k,e} or z_t, both of which carry a temporal/sequence dimension). The roles of N (number of weight matrices) and the reshaping of the CP factors into the attention weight matrices of the MSA/MCA layers are not specified. Without a precise statement of the tensor contraction and how the task-specific factors Q^k and λ^k are associated with task embeddings, the mode approximation mechanism cannot be reproduced or independently verified.
minor comments (6)
  1. [Table I, bottom half] The row label 'SEQL' in the LIBERO-SPATIAL / LIBERO-LONG block should be 'SEQUENTIAL' for consistency with the top half.
  2. [Section IV-A.2, Eqs. (12)-(14)] The definitions of c_{k,k} and c_{q,k} in Eq. (13) omit the epoch index e that appears in the definition of c_{i,j,e}. Please clarify that c_{q,k} is evaluated at e^*_i, following the sentence defining e^*_i, or use an explicit notation.
  3. [Section III-C, after Eq. (6)] The phrase 'the top-C relevant skill weights α_c' should be 'the top-C relevant skills' or 'the weights of the top-C relevant skills'; as written, it suggests the weights are selected independently of the skills.
  4. [Section III-C] The paper does not explain how backward transfer (negative NBT under PACKNET) can improve previously learned tasks when the corresponding skill vectors are frozen. If old skill vectors are frozen, improvement must come from mode-approximation parameters or shared factors; please clarify the mechanism.
  5. [Section IV-A.4] No sensitivity analysis is provided for the hyperparameters M and C (both set to 10). At least one alternative setting (e.g., M=5 or C=5) would help gauge the robustness of the reported results to the codebook size.
  6. [Fig. 4] The color-coded skill provenance in Fig. 4 is difficult to discern in grayscale; please use distinct hatching or patterns in addition to color.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: SPECI's claims are empirical benchmark comparisons against external baselines, and the only flagged gap (an unverified orthogonalization assertion) is a correctness concern, not a circular derivation.

full rationale

No circularity found in this paper. The central claim—that SPECI outperforms state-of-the-art CIL methods on LIBERO—is supported by Tables I and II against external flat-policy baselines, BUDS, LOTUS, ER, and PACKNET, and the metrics FWT, NBT, and AUC are computed from LIBERO success rates rather than being defined in terms of SPECI's fitted skill vectors or CP-decomposition factors. The skill codebook and attention selection (Eqs. 4-6) and mode approximation (Eqs. 10-11) are presented as architectural designs drawn from external cited work ([54] and [56]) with modifications, not as predictions derived from those citations. There are no self-citations and no uniqueness theorem imported from the authors' prior work. The only flagged issue is a missing-support assertion in Section III-C, where the paper states that "applying Schmidt orthogonalization before learning new tasks is enough to prevent interference between key and skill vectors acquired across different tasks"; no proof, ablation, or implementation detail is provided, making this a correctness and verifiability risk rather than a circular step, since no evaluated metric or predicted success rate is defined in terms of that assertion. The labeling of the SEQUENTIAL-SPECI baseline as an "upper bound" is a misnomer, especially because SPECI exceeds it on LIBERO-LONG, but a mislabeled baseline does not make the derivation circular.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The central claim rests on several domain assumptions and a few ad hoc architectural assumptions, the most fragile being that a fixed-size discrete codebook with frozen old entries can represent and compose manipulation skills without interference. No formal guarantees are provided; the paper relies on empirical results. The main free parameters are hyperparameters that control codebook capacity, selection width, and decomposition rank, none of which are fitted to the evaluation data in a circular way.

free parameters (6)
  • M (skill vectors per task) = 10
    Controls the size of each task's skill subset and the total codebook capacity; set empirically in Section IV-A.
  • C (top-C selected skills) = 10
    Number of skill vectors combined at each step; controls the composition width of the synthesized latent skill; set empirically in Section IV-A.
  • R (CP decomposition rank) = 64
    Rank of the mode approximation tensor decomposition; controls the capacity of task-specific attention parameters; set empirically in Section IV-A.
  • N (number of transformer decoder blocks)
    Architecture depth for both skill inference and action execution; referenced in Section III-C but never specified in implementation details.
  • GMM component count
    Number of Gaussian components in the action policy head (Eq. 8 uses R but conflicts with CP rank R); never specified.
  • L (temporal sequence length) = 10 x 5
    Sequence length of multimodal temporal states fed to the transformer; set to 10 timesteps times 5 modalities in Section III-B.
assumptions (7)
  • domain assumption All tasks share the same state and action spaces, transition dynamics and horizon; only initial-state distributions and goal predicates differ.
    Stated in Section III-A problem formulation; enables a single unified policy with language conditioning.
  • domain assumption Expert demonstrations are available and behavioral cloning is a suitable learning objective.
    Optimization objective Eq. 2 assumes demonstrations from each task are given and the negative log-likelihood loss is minimized.
  • domain assumption Pretrained CLIP text encoder and ResNet-18 image encoder provide features that can be adapted to manipulation scenes via FiLM and fine-tuning.
    Section III-B relies on these pretrained models as fixed or fine-tuned backbones.
  • ad hoc to paper A discrete codebook of skill vectors, selected by cosine similarity with attention-modified state embeddings and combined by top-C weighted sum, can express the full range of manipulation skills.
    Core architectural assumption in Section III-C; no theoretical or empirical proof that the codebook is sufficiently expressive.
  • ad hoc to paper Freezing old skill subsets and expanding the codebook linearly prevents catastrophic forgetting at the skill level.
    Section III-C states freezing existing skill subsets 'effectively mitigating catastrophic forgetting at the skill level'; this is a design assumption without formal guarantee.
  • ad hoc to paper CP decomposition of attention tensors into shared and task-specific factors can capture task-specific knowledge without interference.
    Mode approximation in Section III-E borrows from [56]; the paper does not justify expressiveness for continual learning.
  • ad hoc to paper Schmidt orthogonalization before learning new tasks is sufficient to prevent interference between key and skill vectors across tasks.
    Section III-C asserts this; no proof, no ablation, and no statement of whether it was actually applied.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SPECI: Skill Prompts based Hierarchical Continual Imitation Learning for Robot Manipulation." pith.science (2026). https://pith.science/paper/XGLWY2HG

@misc{pith2026250415561,
  author       = {Pith},
  title        = {Pith review of: SPECI: Skill Prompts based Hierarchical Continual Imitation Learning for Robot Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XGLWY2HG}},
  note         = {Machine review of arXiv:2504.15561}
}
read the original abstract

Real-world robot manipulation in dynamic unstructured environments requires lifelong adaptability to evolving objects, scenes and tasks. Traditional imitation learning relies on static training paradigms, which are ill-suited for lifelong adaptation. Although Continual Imitation Learnin (CIL) enables incremental task adaptation while preserving learned knowledge, current CIL methods primarily overlook the intrinsic skill characteristics of robot manipulation or depend on manually defined and rigid skills, leading to suboptimal cross-task knowledge transfer. To address these issues, we propose Skill Prompts-based HiErarchical Continual Imitation Learning (SPECI), a novel end-to-end hierarchical CIL policy architecture for robot manipulation. The SPECI framework consists of a multimodal perception and fusion module for heterogeneous sensory information encoding, a high-level skill inference module for dynamic skill extraction and selection, and a low-level action execution module for precise action generation. To enable efficient knowledge transfer on both skill and task levels, SPECI performs continual implicit skill acquisition and reuse via an expandable skill codebook and an attention-driven skill selection mechanism. Furthermore, we introduce mode approximation to augment the last two modules with task-specific and task-sharing parameters, thereby enhancing task-level knowledge transfer. Extensive experiments on diverse manipulation task suites demonstrate that SPECI consistently outperforms state-of-the-art CIL methods across all evaluated metrics, revealing exceptional bidirectional knowledge transfer and superior overall performance.

Figures

Figures reproduced from arXiv: 2504.15561 by the authors.

Figure 1
Figure 1. (a) Overview of SPECI, which consists of three hierarchical modules: Multimodal Perception, Skill Inference and Action Execution. (b) Illustration [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (a) Framework of the proposed SPECI for robot continual imitation learning. The SPECI architecture consists of three hierarchical levels. The [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Comparison of different policy architectures under ER [58] and PACKNET [59] lifelong learning paradigms, evaluated on LIBERO-OBJECT and [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization of skill vector cross-task reuse in our SPECI under [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Visualization of the FWT and AUC metric gaps between the upper [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. SkillPlug: Unsupervised Skill Mining for Few-Shot Adaptation in Robotic Manipulation

    cs.RO 2026-07 conditional novelty 5.0 of 10

    Unsupervised skill mining with self-supervised compactness, alignment, and disentanglement losses yields a fixed skill library that improves multi-task and few-shot robotic manipulation when plugged into ACT and OpenVLA-OFT.

Reference graph

Works this paper leans on

59 extracted references · 27 canonical work pages · cited by 1 Pith paper

  1. [1]

    Avoiding side effects in complex environments,

    A. Turner, N. Ratzlaff, and P. Tadepalli, “Avoiding side effects in complex environments,” Advances in Neural Information Processing Systems, vol. 33, pp. 21 406–21 415, 2020

    work page 2020

  2. [2]

    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

    work page 2018

  3. [3]

    Learning from humans,

    A. G. Billard, S. Calinon, and R. Dillmann, “Learning from humans,” Springer handbook of robotics , pp. 1995–2014, 2016

    work page 1995

  4. [4]

    Sozil: Self- optimal zero-shot imitation learning,

    P. Hao, T. Lu, S. Cui, J. Wei, Y . Cai, and S. Wang, “Sozil: Self- optimal zero-shot imitation learning,” IEEE Transactions on Cognitive and Developmental Systems , vol. 15, no. 4, pp. 2077–2088, 2021

    work page 2021

  5. [5]

    Combining social and intrinsically motivated learning for multitask robot skill acquisition,

    T. Kulak and S. Calinon, “Combining social and intrinsically motivated learning for multitask robot skill acquisition,” IEEE Transactions on Cognitive and Developmental Systems, vol. 15, no. 2, pp. 385–394, 2021

    work page 2021

  6. [6]

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

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, X. Chen, K. Choroman- ski, T. Ding, D. Driess, A. Dubey, C. Finnet al., “Rt-2: Vision-language- action models transfer web knowledge to robotic control,” arXiv preprint arXiv:2307.15818, 2023

    arXiv 2023

  7. [7]

    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, pp. 627–635

    2011

  8. [8]

    Explain- able hierarchical imitation learning for robotic drink pouring,

    D. Zhang, Q. Li, Y . Zheng, L. Wei, D. Zhang, and Z. Zhang, “Explain- able hierarchical imitation learning for robotic drink pouring,” IEEE Transactions on Automation Science and Engineering , vol. 19, no. 4, pp. 3871–3887, 2021

    work page 2021

Show all 59 references
  1. [9]

    Skill transformer: A mono- lithic policy for mobile manipulation,

    X. Huang, D. Batra, A. Rai, and A. Szot, “Skill transformer: A mono- lithic policy for mobile manipulation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 10 852–10 862

    2023

  2. [10]

    Multi-stage cable routing through hierarchical imitation learning,

    J. Luo, C. Xu, X. Geng, G. Feng, K. Fang, L. Tan, S. Schaal, and S. Levine, “Multi-stage cable routing through hierarchical imitation learning,” IEEE Transactions on Robotics , 2024

    2024

  3. [11]

    Learning object-based state estimators for household robots,

    Y . Du, T. Lozano-Perez, and L. P. Kaelbling, “Learning object-based state estimators for household robots,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2022, pp. 12 558–12 565

    2022

  4. [12]

    Task-driven rein- forcement learning with action primitives for long-horizon manipulation skills,

    H. Wang, H. Zhang, L. Li, Z. Kan, and Y . Song, “Task-driven rein- forcement learning with action primitives for long-horizon manipulation skills,” IEEE Transactions on Cybernetics , 2023

    2023

  5. [13]

    Rt-h: Action hierarchies using language,

    S. Belkhale, T. Ding, T. Xiao, P. Sermanet, Q. Vuong, J. Tompson, Y . Chebotar, D. Dwibedi, and D. Sadigh, “Rt-h: Action hierarchies using language,” arXiv preprint arXiv:2403.01823 , 2024

    2024 arXiv

  6. [14]

    Mimicplay: Long-horizon imitation learning by watching human play,

    C. Wang, L. Fan, J. Sun, R. Zhang, L. Fei-Fei, D. Xu, Y . Zhu, and A. Anandkumar, “Mimicplay: Long-horizon imitation learning by watching human play,” inConference on Robot Learning. PMLR, 2023, pp. 201–221

    2023

  7. [15]

    Xskill: Cross embodiment skill discovery,

    M. Xu, Z. Xu, C. Chi, M. Veloso, and S. Song, “Xskill: Cross embodiment skill discovery,” inConference on Robot Learning. PMLR, 2023, pp. 3536–3555

    2023

  8. [16]

    A novel robust imitation learning framework for complex skills with limited demonstrations,

    W. Wang, C. Zeng, H. Zhan, and C. Yang, “A novel robust imitation learning framework for complex skills with limited demonstrations,” IEEE Transactions on Automation Science and Engineering , 2024

    2024

  9. [17]

    Quest: Self- supervised skill abstractions for learning continuous control,

    A. Mete, H. Xue, A. Wilcox, Y . Chen, and A. Garg, “Quest: Self- supervised skill abstractions for learning continuous control,” Advances in Neural Information Processing Systems , vol. 37, pp. 4062–4089, 2024

    2024

  10. [18]

    Prise: Learning temporal action abstractions as a sequence compression problem,

    R. Zheng, C.-A. Cheng, H. Daum ´e III, F. Huang, and A. Kolobov, “Prise: Learning temporal action abstractions as a sequence compression problem,” arXiv preprint arXiv:2402.10450 , 2024

    2024 arXiv

  11. [19]

    Learning skills from demonstrations: A trend from motion primitives to experience abstraction,

    M. Tavassoli, S. Katyara, M. Pozzi, N. Deshpande, D. G. Caldwell, and D. Prattichizzo, “Learning skills from demonstrations: A trend from motion primitives to experience abstraction,” IEEE Transactions on Cognitive and Developmental Systems , 2023

    2023

  12. [20]

    Catastrophic interference in connec- tionist networks: The sequential learning problem,

    M. McCloskey and N. J. Cohen, “Catastrophic interference in connec- tionist networks: The sequential learning problem,” in Psychology of learning and motivation . Elsevier, 1989, vol. 24, pp. 109–165

    1989

  13. [21]

    A comprehensive survey of continual learning: theory, method and application,

    L. Wang, X. Zhang, H. Su, and J. Zhu, “A comprehensive survey of continual learning: theory, method and application,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2024

    2024

  14. [22]

    Overcoming catastrophic forgetting in neural networks,

    J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Desjardins, A. A. Rusu, K. Milan, J. Quan, T. Ramalho, A. Grabska-Barwinska et al. , “Overcoming catastrophic forgetting in neural networks,” Pro- ceedings of the national academy of sciences , vol. 114, no. 13, pp. 3521...

    2017

  15. [23]

    Expe- rience replay for continual learning,

    D. Rolnick, A. Ahuja, J. Schwarz, T. Lillicrap, and G. Wayne, “Expe- rience replay for continual learning,” Advances in neural information processing systems, vol. 32, 2019

    2019

  16. [24]

    Continual learning with deep generative replay,

    H. Shin, J. K. Lee, J. Kim, and J. Kim, “Continual learning with deep generative replay,” Advances in neural information processing systems , vol. 30, 2017

    2017

  17. [25]

    Gradient episodic memory for continual learning,

    D. Lopez-Paz and M. Ranzato, “Gradient episodic memory for continual learning,” Advances in neural information processing systems , vol. 30, 2017

    2017

  18. [26]

    Dualnet: Continual learning, fast and slow,

    Q. Pham, C. Liu, and S. Hoi, “Dualnet: Continual learning, fast and slow,” Advances in Neural Information Processing Systems , vol. 34, pp. 16 131–16 144, 2021

    2021

  19. [27]

    Class-incremental learning with strong pre-trained models,

    T.-Y . Wu, G. Swaminathan, Z. Li, A. Ravichandran, N. Vasconcelos, R. Bhotika, and S. Soatto, “Class-incremental learning with strong pre-trained models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 9601–9610

    2022

  20. [28]

    Expert gate: Lifelong learning with a network of experts,

    R. Aljundi, P. Chakravarty, and T. Tuytelaars, “Expert gate: Lifelong learning with a network of experts,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 3366– 3375

    2017

  21. [29]

    Continual learning for robotics: Definition, framework, learning strategies, opportunities and challenges,

    T. Lesort, V . Lomonaco, A. Stoian, D. Maltoni, D. Filliat, and N. D ´ıaz- Rodr´ıguez, “Continual learning for robotics: Definition, framework, learning strategies, opportunities and challenges,” Information fusion , vol. 58, pp. 52–68, 2020

    2020

  22. [30]

    Lifelong robot learning,

    S. Thrun and T. M. Mitchell, “Lifelong robot learning,” Robotics and autonomous systems, vol. 15, no. 1-2, pp. 25–46, 1995

    1995

  23. [31]

    M2distill: Multi-modal distillation for lifelong imitation learning,

    K. Roy, A. Dissanayake, B. Tidd, and P. Moghadam, “M2distill: Multi-modal distillation for lifelong imitation learning,” arXiv preprint arXiv:2410.00064, 2024

    2024 arXiv

  24. [32]

    Optimizing robotic manipulation with decision-rwkv: A recurrent sequence modeling approach for lifelong learning,

    Y . Dong, T. Wu, and C. Song, “Optimizing robotic manipulation with decision-rwkv: A recurrent sequence modeling approach for lifelong learning,” Journal of Computing and Information Science in Engineer- ing, vol. 25, no. 3, 2025. 13

    2025

  25. [33]

    Continual learning for anthropomorphic hand grasping,

    W. Li, W. Wei, and P. Wang, “Continual learning for anthropomorphic hand grasping,” IEEE Transactions on Cognitive and Developmental Systems, vol. 16, no. 2, pp. 559–569, 2023

    2023

  26. [34]

    Bellman- consistent pessimism for offline reinforcement learning,

    T. Xie, C.-A. Cheng, N. Jiang, P. Mineiro, and A. Agarwal, “Bellman- consistent pessimism for offline reinforcement learning,” Advances in neural information processing systems , vol. 34, pp. 6683–6694, 2021

    2021

  27. [35]

    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

    2023 arXiv

  28. [36]

    Single-shot learning of stable dynamical systems for long-horizon manipulation tasks,

    A. St-Aubin, A. Abyaneh, and H.-C. Lin, “Single-shot learning of stable dynamical systems for long-horizon manipulation tasks,” arXiv preprint arXiv:2410.01033, 2024

    2024 arXiv

  29. [37]

    Dynamics-aware unsupervised discovery of skills,

    A. Sharma, S. Gu, S. Levine, V . Kumar, and K. Hausman, “Dynamics-aware unsupervised discovery of skills,” arXiv preprint arXiv:1907.01657, 2019

    1907 arXiv

  30. [38]

    Real-time multisensory affordance-based control for adaptive object manipulation,

    V . Chu, R. A. Gutierrez, S. Chernova, and A. L. Thomaz, “Real-time multisensory affordance-based control for adaptive object manipulation,” in 2019 International Conference on Robotics and Automation (ICRA) . IEEE, 2019, pp. 7783–7790

    2019

  31. [39]

    Bottom-up skill discovery from un- segmented demonstrations for long-horizon robot manipulation,

    Y . Zhu, P. Stone, and Y . Zhu, “Bottom-up skill discovery from un- segmented demonstrations for long-horizon robot manipulation,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 4126–4133, 2022

    2022

  32. [40]

    Ddco: Discovery of deep continuous options for robot learning from demonstrations,

    S. Krishnan, R. Fox, I. Stoica, and K. Goldberg, “Ddco: Discovery of deep continuous options for robot learning from demonstrations,” in Conference on robot learning . PMLR, 2017, pp. 418–437

    2017

  33. [41]

    Cril: Continual robot imitation learning via generative and prediction model,

    C. Gao, H. Gao, S. Guo, T. Zhang, and F. Chen, “Cril: Continual robot imitation learning via generative and prediction model,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2021, pp. 6747–5754

    2021

  34. [42]

    Polytask: Learning unified policies through behavior distillation,

    S. Haldar and L. Pinto, “Polytask: Learning unified policies through behavior distillation,” arXiv preprint arXiv:2310.08573 , 2023

    2023 arXiv

  35. [43]

    Tail: Task-specific adapters for imitation learning with large pretrained models,

    Z. Liu, J. Zhang, K. Asadi, Y . Liu, D. Zhao, S. Sabach, and R. Fakoor, “Tail: Task-specific adapters for imitation learning with large pretrained models,” arXiv preprint arXiv:2310.05905 , 2023

    2023 arXiv

  36. [44]

    Libero: Benchmarking knowledge transfer for lifelong robot learning,

    B. Liu, Y . Zhu, C. Gao, Y . Feng, Q. Liu, Y . Zhu, and P. Stone, “Libero: Benchmarking knowledge transfer for lifelong robot learning,” Advances in Neural Information Processing Systems , vol. 36, 2024

    2024

  37. [45]

    Lifelong robot library learning: Bootstrap- ping composable and generalizable skills for embodied control with language models,

    G. Tziafas and H. Kasaei, “Lifelong robot library learning: Bootstrap- ping composable and generalizable skills for embodied control with language models,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 515–522

    2024

  38. [46]

    Lifelong robot learning with human assisted language planners,

    M. Parakh, A. Fong, A. Simeonov, T. Chen, A. Gupta, and P. Agrawal, “Lifelong robot learning with human assisted language planners,” in 2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 523–529

    2024

  39. [47]

    Incremental learning of retrievable skills for efficient continual task adaptation,

    D. Lee, M. Yoo, W. K. Kim, W. Choi, and H. Woo, “Incremental learning of retrievable skills for efficient continual task adaptation,” Advances in Neural Information Processing Systems , vol. 37, pp. 17 286–17 312, 2024

    2024

  40. [48]

    Lotus: Continual imitation learning for robot manipulation through unsupervised skill discovery,

    W. Wan, Y . Zhu, R. Shah, and Y . Zhu, “Lotus: Continual imitation learning for robot manipulation through unsupervised skill discovery,” in 2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 537–544

    2024

  41. [49]

    A framework for behavioural cloning

    M. Bain and C. Sammut, “A framework for behavioural cloning.” in Machine Intelligence 15 , 1995, pp. 103–129

    1995

  42. [50]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning . PMLR, 2021, pp. 8748–8763

    2021

  43. [51]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

    2016

  44. [52]

    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

  45. [53]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,” Advances in Neural Information Processing Systems, 2017

    2017

  46. [54]

    Coda-prompt: Contin- ual decomposed attention-based prompting for rehearsal-free continual learning,

    J. S. Smith, L. Karlinsky, V . Gutta, P. Cascante-Bonilla, D. Kim, A. Arbelle, R. Panda, R. Feris, and Z. Kira, “Coda-prompt: Contin- ual decomposed attention-based prompting for rehearsal-free continual learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision a...

    2023

  47. [55]

    Towards a unified view of parameter-efficient transfer learning,

    J. He, C. Zhou, X. Ma, T. Berg-Kirkpatrick, and G. Neubig, “Towards a unified view of parameter-efficient transfer learning,” arXiv preprint arXiv:2110.04366, 2021

    2021 arXiv

  48. [56]

    Parameter-efficient tuning of large-scale multimodal foundation model,

    H. Wang, X. Yang, J. Chang, D. Jin, J. Sun, S. Zhang, X. Luo, and Q. Tian, “Parameter-efficient tuning of large-scale multimodal foundation model,” Advances in Neural Information Processing Systems, vol. 36, pp. 15 752–15 774, 2023

    2023

  49. [57]

    Some mathematical notes on three-mode factor analysis,

    L. R. Tucker, “Some mathematical notes on three-mode factor analysis,” Psychometrika, vol. 31, no. 3, pp. 279–311, 1966

    1966

  50. [58]

    On tiny episodic memories in continual learning,

    A. Chaudhry, M. Rohrbach, M. Elhoseiny, T. Ajanthan, P. K. Dokania, P. H. Torr, and M. Ranzato, “On tiny episodic memories in continual learning,” arXiv preprint arXiv:1902.10486 , 2019

    1902 arXiv

  51. [59]

    Packnet: Adding multiple tasks to a single network by iterative pruning,

    A. Mallya and S. Lazebnik, “Packnet: Adding multiple tasks to a single network by iterative pruning,” in Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , 2018, pp. 7765–7773

    2018

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.