Offline Reinforcement Learning of High-Quality Behaviors Under Robust Style Alignment

Mathieu Petitbois; R\'emy Portelas; Sylvain Lamprier

REVIEW 3 major objections 5 minor 4 references

Offline Reinforcement Learning of High-Quality Behaviors Under Robust Style Alignment

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Style-conditioned offline reinforcement learning can be built by treating styles as goals, using a gated advantage to add task reward without losing alignment.

desk verdict Strong empirical paper on style-conditioned offline RL with a real but non-fatal theoretical gap in the surrogate style reward; deserves a serious referee. read the letter →

arxiv 2601.22823 v2 pith:5MK3DEVQ submitted 2026-01-30 cs.LG cs.AIcs.RO

Mathieu Petitbois , Rémy Portelas , Sylvain Lamprier This is my paper · ORCID

classification cs.LGcs.AIcs.RO

keywords offlinereinforcementlearningstylealignmentgoal-conditionedRLimplicitQ-learningadvantage-weightedregressiontrajectoryrelabelinglabelingfunctionsstylizedpolicies

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 tries to establish a unified, practical way to learn offline policies that both follow a prescribed behavior style and achieve high task rewards. It defines a style as the label of a subtrajectory under a user-supplied labeling function, which makes styles interpretable, cheap to annotate, and measurable. On top of that definition, the paper proposes SCIQL, an offline algorithm that adapts implicit Q-learning with hindsight relabeling and a new Gated Advantage Weighted Regression mechanism to choose actions that improve task performance only when they are unlikely to hurt style alignment. The authors report that SCIQL achieves the best style alignment across every tested dataset and improves the style-versus-task Pareto hypervolume by 41% to 164% over the strongest baseline.

What carries the argument

The central object is the definition of a behavior style as the label of a subtrajectory under a labeling function lambda, replacing episode-level labels with per-step window labels to handle multi-timescale styles and credit assignment. The load-bearing mechanism is the surrogate Markov style reward r^lambda_D(s,a) = p^lambda_{pi_D}(z|s,a), estimated from the offline dataset, which turns style alignment into a standard MDP objective. SCIQL combines implicit Q-learning for style-conditioned value functions with hindsight relabeling (sampling style labels outside the joint distribution) and a Gated Advantage Weighted Regression update, xi = A^lambda + sigma(A^lambda) * A^r, which approximates

What would settle it

Run SCIQL on a dataset where the label predictor p(z|s,a) is deliberately misaligned with the true window-based style labels (for example, state-action pairs labeled z that never appear near a window labeled z), then measure the empirical alignment S1 on held-out trajectories; if SCIQL achieves high surrogate reward but low true alignment, the transfer assumption is falsified.

Watch

Extended reading notes

Core claim

The central claim is that SCIQL simultaneously achieves superior style alignment and style-conditioned task performance compared to prior offline methods. Style alignment is optimized by replacing the true non-Markovian style objective with a dataset-estimated Markovian surrogate reward, r^lambda_D(s,a) = p^lambda_{pi_D}(z|s,a), then learning style-conditioned value functions through expectile regression with hindsight style relabeling. Task performance is added through Gated Advantage Weighted Regression, which weights behavior cloning by a gated combination of the style advantage and the task advantage, so task improvement is allowed mainly when the style signal is positive. The paper repo

Load-bearing premise

The paper assumes, without proof, that the dataset-estimated Markovian style reward p^lambda_{pi_D}(z|s,a) is an adequate proxy for the true non-Markovian style alignment objective; the authors state this is only supported experimentally, and if optimizing this surrogate does not transfer to the true metric, the central style-alignment claim fails.

Editorial extensions

If this is right

If the surrogate style reward transfers, offline agents can be trained to exhibit user-defined styles from static, heterogeneous datasets without manual trajectory labels or environment interaction.
Hindsight relabeling enables styles that never appear together in a single training trajectory, because the policy can stitch behaviors across different demonstrations.
The gated advantage mechanism allows task-performance improvement without sacrificing style alignment, effectively shifting the Pareto frontier of style-versus-task outcomes toward the ideal point.
SCIQL's alignment is robust to label noise: it maintains better alignment than all baselines under noisy annotations and only converts to following wrong labels past a clearly computable noise threshold.
The framework extends goal-conditioned RL ideas to stylized policies, potentially opening a route to zero-shot style-conditioned control at inference time.

Reading between the lines

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

If the surrogate Markov reward is a faithful proxy for the true window-based style objective, the same framing could be applied to unsupervised or learned style criteria, replacing labeling functions with learned trajectory encoders and retaining the same SCIQL machinery.
The gating idea, using a sigmoid on the style advantage to scale the task signal, is a generic mechanism that could be transferred to other multi-objective offline RL settings, such as safety or constraint satisfaction, wherever one objective must be preserved while a second is optimized.
The paper leaves open the question of multiple simultaneous style criteria; a natural testable extension is whether separate style value functions and gates can be composed, or whether a single criterion needs to be hand-selected.
Because the style reward is estimated from the dataset, the approach may inherit dataset bias: styles that are rare or absent in the data cannot be reliably aligned, so a practical test is whether SCIQL degrades gracefully as label coverage shrinks.

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

Quick take: this is a solid empirical paper that deserves refereeing. The core contribution is SCIQL, which combines IQL with hindsight relabeling and a gated AWR update to optimize style alignment and task performance from a static dataset. The experiments are extensive — 7 datasets, 820 training runs, multiple ablations — and the gains are generally large and consistent across environments, especially on HalfCheetah and HumEnv.

What is genuinely new: the subtrajectory-window labeling function formulation is a useful generalization of the trajectory-level labels in CTV AE and BCPMI, and the GAWR gating mechanism is a simple, effective way to add task reward signals without trashing the style. The paper also includes a careful evaluation protocol and releases code and data, which is more than many papers in this area do.

The soft spots are in the theory. The surrogate Markov reward in Eq. (5) replaces the non-Markovian true objective Eq. (2) without any bound; the paper explicitly says “we show experimentally” that it works. The stress-test concern is legitimate: the implementation actually uses a per-transition indicator (Eq. 12), which is even more myopic, and for window-based labels (turn direction, radius, curvature noise) a learned policy could in principle exploit the surrogate by visiting the same (s,a) in a different temporal context. That said, the evaluation is on the true metric S1 (Eq. 16), not on the surrogate, so the paper is not circular — it just lacks a theoretical safety net. A reviewer should push for at least a characterization of when the surrogate can diverge, or a consistency argument.

The other soft spots are less serious. The GAWR gate is a heuristic; it works empirically, but there is no analysis of what it is doing to the style-optimal set. The estimator choice in Appendix E.1 is made by looking at the final style-alignment numbers, and the global choice of style sampling distribution p_r is not optimal for all criteria (Appendix E.2 shows p_c would help on several Circle2d criteria). These are tunable, but they mean the headline numbers come with some selection-on-metric caveat. Finally, the manuscript has artifacts: a missing citation (“?model” in Related Work), a Table?? reference, and inconsistent project URLs. That suggests the paper was rushed out.

Bottom line: the central claim — that SCIQL beats prior methods on both axes — holds up well in the experiments. The theoretical gap is real but not fatal, because the evaluation targets the true objective. I would send this to peer review, with a request for a stronger theoretical treatment of the surrogate and a cleanup of the manuscript. The paper is worth citing for anyone working on style-conditioned offline RL.

Referee Report

3 major / 5 minor

Summary. The paper proposes a general definition of behavior style as a subtrajectory labeling function, casts style-aligned policy learning as a goal-conditioned offline RL problem, and introduces SCIQL, which combines IQL-style value learning with hindsight relabeling and a new Gated Advantage Weighted Regression (GAWR) rule. The conceptual objective (Eq. 7) is to maximize task return among policies that maximize style alignment. In practice, the true non-Markovian style objective is replaced by a Markovian surrogate reward r_D^λ(s,a)=p^λ_{π_D}(z|s,a), and the policy is extracted via weighted behavior cloning with a gated advantage ξ=Aλ+σ(Aλ)Ar. Experiments on Circle2d, HalfCheetah, and HumEnv across 7 datasets and 820 training runs report large improvements over CBC, BCPMI, SORL, and ablations in both style alignment and style-conditioned task performance.

Significance. If the empirical findings hold, this is a practically valuable contribution: the labeling-function formulation makes styles interpretable, cheap to specify, and directly measurable, while SCIQL demonstrates that value learning plus relabeling can improve style alignment and task performance from static heterogeneous data. The paper is strong on reproducibility: code, datasets, environment details, and per-criterion results are provided, and the evaluation is unusually extensive. However, the core method rests on two unsupported approximations — the Markovian surrogate style reward and the GAWR projection — and the paper's strongest claims are stated in absolute terms. The contribution is therefore primarily empirical; the theory gaps need to be addressed before the claims can be accepted at face value.

major comments (3)

[Sec. 3, Eq. (5); Sec. 4.2, Eq. (12)] The central style-alignment claim rests on an unproven surrogate. Eq. (5) replaces the true objective S1 (Eq. 2) with Sp via r_D^λ(s,a)=p^λ_{π_D}(z|s,a), with no bound or consistency argument; Sec. 3 only promises experimental support. The actual TD target in Eq. (12) is more myopic: χ_ind=1(z=z_c) is a per-transition indicator, and with z sampled uniformly over labels (p_r, App. B), the target is 1 only when the sampled label matches the dataset label; this is not p(z|s,a) in any update, and ignores w>1 temporal dependence. Under distribution shift, a policy can visit the same (s,a) in a different context and receive reward for the wrong window label. Table 3 shows symptoms (Circle2d movement direction: 20.5–27.0 alignment, vs. near-100 for position). A formal bound or a direct Sp-vs-S1 transfer experiment is required to support 'best style alignment by a large margin.'
[Sec. 4.3, Eq. (14)] GAWR is presented as approximating the projection onto the style-optimal set Π_λ(z) of Eq. (7), but no derivation, convergence result, or error bound is given. The gated combination ξ=Aλ+σ(Aλ)Ar is a heuristic whose sigmoid makes it scale-dependent; the EMA normalization mentioned after Eq. (14) is not specified in Appendix B. Table 4 shows the gate can hurt style: halfcheetah-stitch-v0 style drops from 78.0 (SCIQL λ) to 60.8 (SCIQL λ>r), and several Circle2d criteria collapse under r>λ. Since the task-performance claims depend on this rule, the paper should either characterize the implied constraint formally (e.g., a style-return lower bound under a margin condition) or compare systematically against a constrained-RL baseline and an unweighted sum ablation.
[Sec. 5.3, Fig. 2] The hypervolume claims are not fully supported. SORL is evaluated at only β=0,1,3 and SCIQL at only λ, λ>r, r>λ; these six points do not define a Pareto front, so the reported '+41.2% to +163.9%' hypervolume improvement is an artifact of the chosen configurations. The text also reports an '18–28%' reduction in Euclidean distance to the ideal point without giving the formula, the reference point, or confidence intervals. Please report the exact hypervolume definition, the set of points used, and seed-level variability, or compute a true Pareto hull over all evaluated trade-off settings.

minor comments (5)

[Sec. 2, Related Work] There is a missing reference placeholder: 'Also in the online RL setting, ?model the stylized policy learning problem...' should cite the relevant constrained-MDP/online stylized-RL papers.
[Sec. 5.2 / Table 3 / App. D] The main-text claim 'best style alignment performance by a large margin compared to previous baselines for every dataset' is true for the averaged dataset scores, but Table 3 shows SCIQL is below CBC/BCPMI on Circle2d turn direction, radius, and speed. This is acknowledged in Appendix D, but the main text should qualify 'average across criteria' more visibly.
[Algorithm 1 / Sec. 4.3] Algorithm 1 lists three sequential 'while not converged' loops, while the text says value learning and policy extraction 'can be performed jointly in practice.' Clarify which training scheme was actually used for the reported results.
[Appendix B] The GAWR normalization ('advantages can be normalized with an exponential moving average') is not specified — EMA decay, normalization target, and whether it is applied to Aλ, Ar, or both should be given.
[Various] Table 4 caption and the accompanying text refer to 'Table??' instead of the actual table number. Several references are incomplete or duplicated in the bibliography; a final proofread is needed.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: SCIQL is evaluated against an external window-label metric; the surrogate-objective gap and GAWR heuristic are unproven approximations, not circular reductions.

full rationale

I walked the derivation chain from the true style objective S1 (Eq. 2) through the probabilistic relaxation S~p (Eq. 4), the dataset surrogate Sp (Eq. 5), the SCIQL Q-learning losses (Eqs. 11-12), the AWR policy extraction (Eqs. 13, 15), and the reported evaluations (Eq. 16, Tables 1, 3-5, Figure 2). The central style-alignment and task-performance claims are not derived by fitting the target result. Training uses a dataset-derived Markov reward (Eq. 5; implemented as chi=1(z=z_c) in Eq. 12), while evaluation computes the empirical non-Markovian window-label alignment S1 (Eq. 16) on fresh rollouts. These are not equal by construction: the same state-action pair can appear under different temporal contexts, and the paper's own hard-coded labels (turn direction, radius, curvature noise) use windows w>1, so the surrogate and the metric can diverge. That gap is a correctness risk, explicitly acknowledged by the paper in the sentence "We show experimentally that this objective still permits high alignment when computing the empirical true style alignment metric in Equation 16" (Sec. 3) — no bound is provided. Similarly, GAWR (Eq. 14) is described as "a practical approximation to the projection in Equation 7" (Sec. 4.3) without a formal proof; this is an unproven heuristic, not a circular step. Minor self-citation exists: contribution (3) credits "style-relabeling (Petitbois et al., 2025)" (Sec. 1), and Related Work describes WZBC (Petitbois et al., 2025). This is not load-bearing because the hindsight-relabeling mechanism is grounded in HER (Andrychowicz et al., 2017) and IQL (Kostrikov et al., 2021), which are external standard references. One mild validation concern is that Appendix E.1 selects the chi estimator by comparing style-alignment scores ("in our experiments we adopt softmax for SORL and ind for SCIQL"), which is model selection on the evaluation metric rather than a construction-level fit; it does not force the reported rollouts. Overall, the algorithm is benchmarked against external baselines and its predictions are not equivalent to its inputs by definition.

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

The central method depends on standard IQL assumptions, on the unproven surrogate Markov style reward, and on the heuristic GAWR gate. The main free choices are the IQL hyperparameters, the style-label sampling distribution, and the χ estimator, the last two selected on the evaluation metric in appendices.

free parameters (5)

Expectile κ = 0.7
Standard IQL hyperparameter, chosen by hand; not fitted to style results.
AWR inverse temperature β = 3.0
Standard IQL/AWR temperature; controls policy sharpness.
Style-label sampling distribution p_m = p_r (uniform over labels) for main results
Global choice; Appendix E.2 shows p_c improves Circle2d turn/radius/speed alignment, so this choice changes reported averages.
χ estimator for p(z|s,a) = Indicator of current label for SCIQL; softmax for SORL
Selected empirically using style-alignment results on the same benchmarks (Table 5, Fig. 17).
Label window sizes / bin edges = w=1 for most criteria; 11/51 for turn/radius/noise; user-specified thresholds
Defines the style labels themselves; chosen by hand, not learned.

assumptions (5)

domain assumption The offline dataset D has sufficient coverage and the behavior policy π_D is such that IQL value estimates are reliable.
Standard offline RL assumption; used in Section 4.1 and Section 4.2.
ad hoc to paper The Markovian surrogate style reward p^λ_{π_D}(z|s,a) is an adequate proxy for the true non-Markovian style alignment objective.
Section 3, Equations (4)-(5); no bound or proof, only experimental evidence.
ad hoc to paper The gated combination ξ = A_λ + σ(A_λ) A_r approximates the projection onto the style-optimal set Π_λ(z).
Section 4.3 Eq. (14); presented as a practical approximation with no formal guarantee.
standard math Expectile regression and AWR converge in the offline setting as in IQL.
Relies on Kostrikov et al. (2021) and Peng et al. (2019).
domain assumption The labeling function λ defines the ground-truth style; no independent style metric is used for evaluation.
The evaluation metric in Eq. (16) uses the same λ as training; inherent to the setting but limits external validity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Offline Reinforcement Learning of High-Quality Behaviors Under Robust Style Alignment." pith.science (2026). https://pith.science/paper/5MK3DEVQ

@misc{pith2026260122823,
  author       = {Pith},
  title        = {Pith review of: Offline Reinforcement Learning of High-Quality Behaviors Under Robust Style Alignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5MK3DEVQ}},
  note         = {Machine review of arXiv:2601.22823}
}

read the original abstract

We study offline reinforcement learning of style-conditioned policies using explicit style supervision via subtrajectory labeling functions. In this setting, aligning style with high task performance is particularly challenging due to distribution shift and inherent conflicts between style and reward. Existing methods, despite introducing numerous definitions of style, often fail to reconcile these objectives effectively. To address these challenges, we propose a unified definition of behavior style and instantiate it into a practical framework. Building on this, we introduce Style-Conditioned Implicit Q-Learning (SCIQL), which leverages offline goal-conditioned RL techniques, such as hindsight relabeling and value learning, and combine it with a new Gated Advantage Weighted Regression mechanism to efficiently optimize task performance while preserving style alignment. Experiments demonstrate that SCIQL achieves superior performance on both objectives compared to prior offline methods. Code, datasets and visuals are available in: https://mathieu-petitbois.github.io/projects/sciql/.

Figures

Figures reproduced from arXiv: 2601.22823 by the authors.

**Figure 1.** Long term decision making and stitching challenges for style alignment optimization. Achieving movement styles such as high-speed running may require to standing and accelerating, which means navigating through different speed styles and demands long-term decision making. Also, trajectories in D may not cover all the speed styles, calling for trajectory stitching such as (slow → medium) and (medium → fast). Challeng… view at source ↗

**Figure 2.** Pareto fronts and hypervolumes of SORL and SCIQL. We compare SORL (in blue) and SCIQL (in red). The shaded areas ( , ) represent the hypervolumes covered by the methods. Markers indicate different trade-off configurations: SORL is evaluated at β = 0 ( ), β = 1 ( ), and β = 3 ( ). SCIQL is evaluated with style only λ ( ), style-prioritized λ > r ( ), and task-prioritized r > λ ( ). SCIQL consistently achieves a large… view at source ↗

**Figure 3.** Circle2d environment visualizations. Environment The Circle2d environment consists of a 2D plane where an agent can roam around within a confined square. Its state space S corresponds to the history of the previous 4 (xagent, yagent, θagent) ∈ [[xmin, xmax] × [ymin, ymax] × [θmin, θmax]] = [−50.0, 50.0] × [−50.0, 50.0] × [−π, π], padded if needed by repeating the oldest triplet (namely for the beginning of the traj… view at source ↗

Figures from the paper (16 more)

**Figure 4.** Figure 4: Circle2d datasets trajectory visualizations at different percentages. The top row corresponds to the circle2d-inplace-v0 while the bottom row corresponds to the circle2d-navigate-v0 Criteria and labels We present below the various labeling functions we designed for Ci…

**Figure 5.** Figure 5: Circle2d position label visualizations at different percentages. (a) inplace - 5% (b) inplace - 100% (c) navigate - 5% (d) navigate - 100% [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]

**Figure 6.** Figure 6: Circle2d movement direction label visualizations at different percentages. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]

**Figure 7.** Figure 7: Circle2d turn direction label visualizations at different percentages. (a) inplace - 5% (b) inplace - 100% (c) navigate - 5% (d) navigate - 100% [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]

**Figure 8.** Figure 8: Circle2d radius label visualizations at different percentages. (a) inplace - 5% (b) inplace - 100% (c) navigate - 5% (d) navigate - 100% [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]

**Figure 9.** Figure 9: Circle2d speed label visualizations at different percentages. A.2 HALFCHEETAH Environment HalfCheetah (Todorov et al., 2012; Towers et al., 2024) is an environment consisting of controlling a 6-DoF 2-dimensional robot composed of 9 body parts and 8 joints connecting t…

**Figure 10.** Figure 10: Circle2d curvature noise visualizations at different percentages. Task As implemented in Towers et al. (2024), at each timestep t, the agent applies continuous control actions at ∈ R d that drive the joints of the cheetah. The environment evaluates performance using a…

**Figure 12.** Figure 12: • torso height: The torso–height labeling function λtorso discretizes the vertical torso position ht. We define [hmin, hmax] = [0.4, 0.8] (real units) and split into K = 3 bins, giving L(λtorso) = J0, 2K. Labels are assigned per timestep and smoothed by majority vote.…

**Figure 11.** Figure 11: HalfCheetah speed label histograms. (a) halfcheetah-fix-v0 (b) halfcheetah-stitch-v0 (c) halfcheetah-vary-v0 [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]

**Figure 13.** Figure 13: HalfCheetah torso height label histograms. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_13.png]

**Figure 14.** Figure 14: HalfCheetah backfoot height label histograms. (a) halfcheetah-fix-v0 (b) halfcheetah-stitch-v0 (c) halfcheetah-vary-v0 [PITH_FULL_IMAGE:figures/full_fig_p022_14.png]

**Figure 15.** Figure 15: HalfCheetah frontfoot height label histograms. A.3 HUMENV Environment The HumEnv environment (Tirinzoni et al., 2025) is built on the SMPL skeleton (Loper et al., 2023), which consists of 24 rigid bodies, among which 23 are actuated. This SMPL skeleton is widely used …

**Figure 16.** Figure 16: c. 0 1 label 0.0 0.1 0.2 0.3 0.4 0.5 proportion (a) simple - head height 0 1 2 label 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 proportion (b) complex - speed 0 1 2 label 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 proportion (c) complex - head height [PITH_FULL_I…

**Figure 17.** Figure 17: Style alignments histograms for different p(z|s, a) estimation strategies. E ABLATIONS For the following, we define the dataset joint sampling distribution: p λ(D) (z|s, a) and a mixture p λ(D) m (z|s, a) of a set of sampling distributions: p λ(D) c (z|s, a) which cor…

**Figure 18.** Figure 18: SCIQL performance under p λ(D) r vs p λ(D) c ? 32 [PITH_FULL_IMAGE:figures/full_fig_p032_18.png]

**Figure 19.** Figure 19: Evolution of style alignment under noisy labels. For noise labels ζ ∈ {0.0, 0.1, ..., 1.0}, we compare the evolution of style alignment of BC ( ), CBC ( • ), BC-PMI ( • ), SCBC ( • ), SORL ( • ) and SCIQL ( • ). We see that SCIQL maintains an overall better alignment…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

4 extracted references

[1]

write newline

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

@esa (Ref

\@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...
[3]

\@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...
[4]

the vector of an unsupervised learned trajectory encoder

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

arXiv 2026

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

write newline

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

[2] [2]

@esa (Ref

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

[3] [3]

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

[4] [4]

the vector of an unsupervised learned trajectory encoder

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

arXiv 2026