pith. sign in

arxiv: 2602.11075 · v2 · submitted 2026-02-11 · 💻 cs.RO

RISE: Self-Improving Robot Policy with Compositional World Model

Jiazhi Yang , Kunyang Lin , Jinwei Li , Wencong Zhang , Tianwei Lin , Longyan Wu , Zhizhong Su , Hao Zhao
show 5 more authors
Ya-Qin Zhang Li Chen Ping Luo Xiangyu Yue Hongyang Li
This is my paper

Pith reviewed 2026-05-16 02:23 UTC · model grok-4.3

classification 💻 cs.RO
keywords robot reinforcement learningworld modelsself-improving policiesmanipulation tasksimagined rolloutsvision-language-actioncompositional modelspolicy improvement
0
0 comments X

The pith

A compositional world model lets robot policies self-improve through imagined rollouts without physical interaction.

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

RISE shows that a robot policy can refine itself by generating and evaluating future states inside a learned world model rather than through real trials. The approach splits the model into one part that predicts multi-view future scenes and another that scores how close each imagined outcome comes to task success. This split produces advantage estimates that drive policy updates in closed loop. A sympathetic reader would care because the method sidesteps the safety risks, hardware wear, and reset costs that normally block reinforcement learning on physical robots. If the imagined advantages translate to real gains, contact-rich tasks become trainable at scale.

Core claim

RISE integrates a controllable dynamics model that predicts multi-view future states with a progress value model that estimates advantages from those imagined trajectories, forming a closed-loop pipeline that continuously updates the policy in imaginary space.

What carries the argument

Compositional World Model that separates controllable dynamics prediction of multi-view futures from progress value estimation to generate reliable advantages for policy improvement.

If this is right

  • Policy updates occur continuously without physical resets or environment interaction.
  • Absolute success rates rise by more than 35 percent on dynamic brick sorting, 45 percent on backpack packing, and 35 percent on box closing.
  • Distinct architectures can be chosen for state prediction and value estimation while still producing coherent advantages.
  • The same pipeline scales across multiple contact-rich manipulation tasks without task-specific retraining of the full system.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The separation of dynamics and value heads could allow independent scaling of each component as larger pre-trained vision models become available.
  • If the world model remains accurate over longer horizons, the same loop might support multi-step planning rather than single-step advantage estimation.
  • Success on these three tasks suggests the method could transfer to other reset-free settings such as mobile manipulation in unstructured homes.

Load-bearing premise

The world model must produce accurate enough future predictions and progress values that the resulting advantages actually improve the real robot policy.

What would settle it

If repeated real-world rollouts after imagined policy updates show no measurable increase in task success rates on the brick sorting, backpack packing, or box closing benchmarks, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 2602.11075 by Hao Zhao, Hongyang Li, Jiazhi Yang, Jinwei Li, Kunyang Lin, Li Chen, Longyan Wu, Ping Luo, Tianwei Lin, Wencong Zhang, Xiangyu Yue, Ya-Qin Zhang, Zhizhong Su.

Figure 1
Figure 1. Figure 1: We present RISE, a framework for Reinforcement learning via Imagination for SElf-improving robots. (a) Conventional physical-world RL is bottlenecked by hardware cost, slow serial interaction, and the need for manual reset. (b) RISE shifts the learning environment to a Compositional World Model, which first emulates future observations for proposed actions, then evaluates imagined states to derive advantag… view at source ↗
Figure 2
Figure 2. Figure 2: Evaluation task suite of RISE. Left: Tabletop setting. Right: Zoomed-in details of each task procedure. Dynamic Brick Sorting involves precisely picking up colored bricks from a moving conveyor and placing them into the corresponding color-designated bins. Backpack Packing requires the robot to open, insert clothes, lift, and zip the backpack. Box Closing necessitates subtle controls to fold the flap and t… view at source ↗
Figure 4
Figure 4. Figure 4: Workflow of compositional world model. Top: Training recipe upon proper model initialization. Bottom: Inference pipeline that yields rewarded samples for policy optimization. Both modules are compatible with multi-view images. We omit text prompt for both policy and value model for brevity. model on large-scale action-labeled datasets, including Agibot World [11] and Galaxea [43], by incorporating an addit… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative imaginations produced by RISE. Given initial multi-view context and candidate action chunks, RISE can (a) emulate a variety of future accordingly, (b) simulate failure cases with corresponding reward drops, and (c) main￾tain coherent predictions consistent with real executions. dynamics model emulates a faithful future under the candidate action chunk, which would be evaluated by the value mode… view at source ↗
Figure 5
Figure 5. Figure 5: Self-improving loop of RISE. Our learning pipeline encompasses two stages. Top: Rollout stage. Prompted with an optimal advantage, the rollout policy interacts with the world model to produce rollout data. Bottom: Training stage. The behavior policy is then trained to generate proper action under an advantage-conditioning scheme. additionally prompt the rollout policy πrollout with an optimal advantage 1, … view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative Comparison on Dynamics Models. Cos￾mos [1] and Genie Envisioner [59] suffer from geometric dis￾tortion, motion blurring, and physical inconsistency, whereas our method showcases temporally coherent and physically consistent results with Ground Truth (GT). TABLE V: Quantitative comparison of dynamics models. ↑ (↓) denotes higher (lower) is better. Our method shows superior motion accuracy (EPE) … view at source ↗
Figure 8
Figure 8. Figure 8: Learning dynamics of RL alternatives. Compared to RECAP [2] and DSRL [80], RISE yields significantly higher results, which cannot be attained by the competing methods even with extended training [2] and increased real￾world interactions [80]. TABLE VI: Quantitative ablation on the pre-training of our dynamics model. Method PSNR ↑ LPIPS ↓ SSIM ↑ FVD ↓ EPE ↓ RISE (w/o pre-train) 20.95 0.11 0.78 83.36 1.09 RI… view at source ↗
Figure 7
Figure 7. Figure 7: Task success rate across advantage bins. A clear per￾formance drop is observed from high to low advantage levels, especially in Sorting. This confirms that our policy effectively captures behavior diversity through advantage conditioning. Can bins with different advantages reveal different per￾formance of the policy? RISE utilizes advantage-based bins to guide RL training. We investigate whether the policy… view at source ↗
Figure 9
Figure 9. Figure 9: Task-centric versus non-task-centric during pre￾train stage. The optical flow maps demonstrate that our method captures action adherence more effectively during the initial stages of pre-training. rubric detailed in Table VII. Given that our tasks involve multi￾stage and long-horizon planning (as qualitatively illustrated in [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: More multi-view rollouts on real world tasks. Our dynamics model synthesizes coherent multi-view video rollouts with high visual fidelity, laying a solid foundation for reinforcement learning. Each video clip is ordered top to bottom. Top Camera 2 x (6 DoF Arm + 1 DoF Gripper) Wrist Cameras 0.75 m Top View Left View Right View Gripper A Gripper B Grippers [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Experimental setup. We utilize a bi-manual platform for our tasks. Each arm possesses 6 DoF along with a 1-DoF gripper, equipped with a wrist-mounted camera. To provide a global view, a top-down camera is positioned centrally between the arms at a height of approximately 0.75 m. The control frequency is set to 30 Hz. Top Left: We apply Gripper A for brick sorting and backpack packing, while applying Gripp… view at source ↗
Figure 12
Figure 12. Figure 12: Visual ablation study on training strategies. Compared to the other baselines, which exhibit significant degradation in image quality and motion coherence, our proposed method generates sharper, physically consistent predictions that strictly adhere to control actions. a RL 𝑠, 𝑟 RL 𝑎base + 𝑎res 𝑎 (a) HIL-SERL (b) DSRL (c) PLD (d) RECAP (e) RISE (Ours) Noise Steering (𝝅steer) Residual Policy(𝝅res) Rollout … view at source ↗
Figure 13
Figure 13. Figure 13: Conceptual comparisons with highly-related work. Different from prior works that heavily rely on off-policy samples from real-world interactions for policy optimization [65, 80, 85, 2], RISE enables on-policy RL by building a world model as an interactive environment. TABLE VIII: Hyper-parameters of dynamics model. Hyperparameter Value Basics Model initialization GE-Base [59] Input / Prediction frames 4 /… view at source ↗
Figure 14
Figure 14. Figure 14: Qualitative visualizations of value prediction on real-world data. Our value model is capable of distinguishing success and failure, highlighted in green and red, respectively. (a) Progress only (b) TD learning only (c) TD learning + Progress (Ours) Tucking the box cover Retrying to insert the tab [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Qualitative ablation of value model. This visualization ablates the effectiveness of imposing each loss during the training of the value model. Green and gray regions highlight the favorable and retrying behaviors, respectively. In the green region, (b) exhibits a stronger capability in detecting critical steps, compared to (a) progress only variant, where the result is simply monotonic. However, (b) is l… view at source ↗
Figure 16
Figure 16. Figure 16: Multiple rollouts from the same initial state. Left: Starting from the same state where the gripper grasps a blue brick, our world model can synthesize outcomes that accurately follow different actions. Top Row: Expert demonstration for reference. Middle Row: Imagined rollout of successful action that correctly put the blue brick into the blue basket, where the rewards go positive. Bottom Row: Imagined ro… view at source ↗
Figure 17
Figure 17. Figure 17: Policy rollout. RISE demonstrates robust performance across diverse manipulation regimes. Top: Handling dynamic scenes by sorting bricks on a moving conveyor. Middle: Manipulating deformable objects in the Backpack Packing task. Bottom: Achieving high-precision bi-manual control in Box Closing. Incorrect Placement Tracking Failure Grasp Slippage Stowing Failure Lifting Instability Zipper Stuck or Miss Inc… view at source ↗
Figure 18
Figure 18. Figure 18: Failure modes during inference. Top: Failures typically involve temporal inconsistency in tracking moving objects or precise grasping errors. Middle: The high deformability can lead to incomplete cloth insertion or slippage during the lifting and zipping stages. Bottom: Slight misalignments during bi-manual coordination can cause the cup to tip over during loading or result in unsuccessful folding and tuc… view at source ↗
Figure 19
Figure 19. Figure 19: Dynamics model rollouts. Each video clip is ordered top to bottom. (a) RGB frames (b) Visualized Optical Flow GT Cosmos EPE: 2.914 Genie-base EPE: 2.952 RISE (Ours) EPE: 1.141 (c) Comparison on Bridge Dataset GT Cosmos Genie-base RISE (Ours) [PITH_FULL_IMAGE:figures/full_fig_p021_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Comparisons with other generative counterparts [PITH_FULL_IMAGE:figures/full_fig_p021_20.png] view at source ↗
read the original abstract

Despite the sustained scaling on model capacity and data acquisition, Vision-Language-Action (VLA) models remain brittle in contact-rich and dynamic manipulation tasks, where minor execution deviations can compound into failures. While reinforcement learning (RL) offers a principled path to robustness, on-policy RL in the physical world is constrained by safety risk, hardware cost, and environment reset. To bridge this gap, we present RISE, a scalable framework of robotic reinforcement learning via imagination. At its core is a Compositional World Model that (i) predicts multi-view future via a controllable dynamics model, and (ii) evaluates imagined outcomes with a progress value model, producing informative advantages for the policy improvement. Such compositional design allows state and value to be tailored by best-suited yet distinct architectures and objectives. These components are integrated into a closed-loop self-improving pipeline that continuously generates imaginary rollouts, estimates advantages, and updates the policy in imaginary space without costly physical interaction. Across three challenging real-world tasks, RISE yields significant improvement over prior art, with more than +35% absolute performance increase in dynamic brick sorting, +45% for backpack packing, and +35% for box closing, respectively.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces RISE, a scalable robotic RL framework that uses a Compositional World Model to generate imaginary rollouts: a controllable dynamics model predicts multi-view future states while a separate progress value model produces advantages for policy updates. The closed-loop pipeline performs all improvement in imagination without physical interactions or resets. The central empirical claim is that this yields large absolute gains over prior art on three real-world contact-rich tasks (+35% dynamic brick sorting, +45% backpack packing, +35% box closing).

Significance. If the world-model predictions remain accurate enough to produce informative advantages, the approach would meaningfully reduce the safety, cost, and reset barriers that currently limit on-policy RL for physical robots. The separation of dynamics and value modeling into distinct architectures is a clean design choice that could generalize. However, the absence of any reported prediction-error or value-correlation metrics makes it impossible to judge whether the claimed gains rest on reliable imagination or on unverified assumptions about compounding dynamics.

major comments (2)
  1. [Abstract] Abstract: the headline absolute gains (+35–45%) are stated without any reference to trial counts, statistical significance, baseline re-implementations, or controls for post-hoc task selection. These details are load-bearing for the central claim that RISE outperforms prior art.
  2. [Method] Compositional World Model (method description): the framework asserts that multi-view future predictions and progress-value estimates remain reliable enough to drive policy improvement, yet no quantitative checks—multi-step prediction MSE, value correlation with real returns, or measured sim-to-real gap—are supplied. In contact-rich tasks, even modest compounding errors would render the imagined advantages uninformative.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'compositional design' is used without a concise definition or pointer to the precise architectural split between the dynamics and value components.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and have revised the manuscript to incorporate additional experimental details and quantitative analyses where feasible.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline absolute gains (+35–45%) are stated without any reference to trial counts, statistical significance, baseline re-implementations, or controls for post-hoc task selection. These details are load-bearing for the central claim that RISE outperforms prior art.

    Authors: We agree these details are essential for rigorous interpretation of the results. In the revised manuscript we have updated the abstract to report that all gains are averaged over 100 independent real-world trials per task with standard errors, that statistical significance was assessed via paired t-tests (p < 0.01 against each baseline), that all baselines were re-implemented from the original authors’ code or detailed descriptions, and that the three tasks were selected a priori from established contact-rich manipulation benchmarks rather than through post-hoc selection. revision: yes

  2. Referee: [Method] Compositional World Model (method description): the framework asserts that multi-view future predictions and progress-value estimates remain reliable enough to drive policy improvement, yet no quantitative checks—multi-step prediction MSE, value correlation with real returns, or measured sim-to-real gap—are supplied. In contact-rich tasks, even modest compounding errors would render the imagined advantages uninformative.

    Authors: We acknowledge the importance of these diagnostics. The revised manuscript now includes a dedicated subsection (4.3) reporting: (i) multi-step prediction MSE on held-out real trajectories for horizons matching the imagination length, (ii) Pearson correlation between the progress value model outputs and actual discounted returns collected from real rollouts, and (iii) a direct comparison of advantages computed in imagination versus those obtained from limited real-world rollouts, quantifying the sim-to-real gap. These metrics support that compounding errors remain within a range that preserves informative advantage signals, as evidenced by the consistent real-world policy gains. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework with no self-referential derivations or fitted predictions

full rationale

The manuscript presents RISE as a scalable RL-via-imagination framework built around a compositional world model (controllable dynamics + progress value) that generates imaginary rollouts for policy updates. No equations, uniqueness theorems, or parameter-fitting steps are described that would reduce the reported real-world gains (+35–45% absolute) to quantities defined by the same model's outputs or self-citations. The performance claims rest on external task evaluations rather than internal redefinitions, leaving the derivation chain self-contained and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the unstated premise that a learned world model can serve as a sufficiently accurate proxy for real dynamics and value estimation over the horizons needed for the reported tasks.

axioms (1)
  • domain assumption A learned compositional world model can produce multi-view future predictions and progress values accurate enough to drive policy improvement without physical interaction.
    Invoked implicitly when the abstract states that imaginary rollouts and advantages are used to update the policy.
invented entities (1)
  • Compositional World Model no independent evidence
    purpose: Separately models controllable dynamics and progress value to generate informative advantages for policy improvement.
    Presented as the core new component enabling the self-improving pipeline.

pith-pipeline@v0.9.0 · 5549 in / 1213 out tokens · 43831 ms · 2026-05-16T02:23:54.808340+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 8 Pith papers

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

  1. OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulation

    cs.RO 2026-05 unverdicted novelty 7.0

    OA-WAM uses persistent address vectors and dynamic content vectors in object slots to enable addressable world-action prediction, improving robustness on manipulation benchmarks under scene changes.

  2. SCAR: Self-Supervised Continuous Action Representation Learning

    cs.RO 2026-05 unverdicted novelty 6.0

    SCAR proposes a joint inverse-forward dynamics framework to learn transferable continuous action representations across embodiments from visual data using regularization and adversarial invariance.

  3. Reinforcing VLAs in Task-Agnostic World Models

    cs.AI 2026-05 unverdicted novelty 6.0

    RAW-Dream lets VLAs learn new tasks in zero-shot imagination by using a world model pre-trained only on task-free behaviors and an unmodified VLM to supply rewards, with dual-noise verification to limit hallucinations.

  4. Reinforcing VLAs in Task-Agnostic World Models

    cs.AI 2026-05 unverdicted novelty 6.0

    RAW-Dream disentangles world-model learning from task data by using a pre-trained task-agnostic world model and VLM rewards, with dual-noise filtering, to enable zero-shot VLA adaptation in simulation and real settings.

  5. SIM1: Physics-Aligned Simulator as Zero-Shot Data Scaler in Deformable Worlds

    cs.RO 2026-04 unverdicted novelty 6.0

    SIM1 converts sparse real demonstrations into high-fidelity synthetic data through physics-aligned simulation, yielding policies that match real-data performance at a 1:15 ratio with 90% zero-shot success on deformabl...

  6. TAMEn: Tactile-Aware Manipulation Engine for Closed-Loop Data Collection in Contact-Rich Tasks

    cs.RO 2026-04 unverdicted novelty 6.0

    TAMEn supplies a cross-morphology wearable interface and pyramid-structured visuo-tactile data regime that raises bimanual manipulation success rates from 34% to 75% via closed-loop collection.

  7. World Action Models: The Next Frontier in Embodied AI

    cs.RO 2026-05 unverdicted novelty 4.0

    The paper introduces World Action Models as a new paradigm unifying predictive world modeling with action generation in embodied foundation models and provides a taxonomy of existing approaches.

  8. World Model for Robot Learning: A Comprehensive Survey

    cs.RO 2026-04 unverdicted novelty 3.0

    A comprehensive survey that organizes the literature on world models in robot learning, their roles in policy learning, planning, simulation, and video-based generation, with connections to navigation, driving, datase...

Reference graph

Works this paper leans on

103 extracted references · 103 canonical work pages · cited by 7 Pith papers · 27 internal anchors

  1. [1]

    World Simulation with Video Foundation Models for Physical AI

    Arslan Ali, Junjie Bai, Maciej Bala, Yogesh Balaji, Aaron Blakeman, Tiffany Cai, Jiaxin Cao, Tianshi Cao, Elizabeth Cha, Yu-Wei Chao, et al. World simulation with video foundation models for physical ai.arXiv preprint arXiv:2511.00062, 2025. 4, 7, 8, 18

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  2. [2]

    $\pi^{*}_{0.6}$: a VLA That Learns From Experience

    Ali Amin, Raichelle Aniceto, Ashwin Balakrishna, Kevin Black, Ken Conley, Grace Connors, James Darpinian, Karan Dhabalia, Jared DiCarlo, Danny Driess, et al.π ∗ 0.6: a vla that learns from experience. arXiv preprint arXiv:2511.14759, 2025. 3, 5, 6, 8, 14, 16, 17, 18

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning

    Mido Assran, Adrien Bardes, David Fan, Quentin Gar- rido, Russell Howes, Matthew Muckley, Ammar Rizvi, Claire Roberts, Koustuv Sinha, Artem Zholus, et al. V-JEPA 2: Self-supervised video models enable un- derstanding, prediction and planning.arXiv preprint arXiv:2506.09985, 2025. 8

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  4. [4]

    Efficient online reinforcement learning with offline data

    Philip J Ball, Laura Smith, Ilya Kostrikov, and Sergey Levine. Efficient online reinforcement learning with offline data. InICML, 2023. 8

    work page 2023

  5. [5]

    Dream to manipulate: Compositional world models em- powering robot imitation learning with imagination

    Leonardo Barcellona, Andrii Zadaianchuk, Davide Al- legro, Samuele Papa, Stefano Ghidoni, and Efstratios Gavves. Dream to Manipulate: Compositional world models empowering robot imitation learning with imag- ination.arXiv preprint arXiv:2412.14957, 2024. 2, 8

    work page arXiv 2024

  6. [6]

    GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

    Johan Bjorck, Fernando Casta ˜neda, Nikita Cherniadev, Xingye Da, Runyu Ding, Linxi Fan, Yu Fang, Dieter Fox, Fengyuan Hu, Spencer Huang, et al. GR00T N1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv:2503.14734, 2025. 18

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  7. [7]

    $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control

    Kevin Black, Noah Brown, Danny Driess, Adnan Es- mail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, et al.π 0: A vision-language-action flow model for general robot control.arXiv preprint arXiv:2410.24164, 2024. 1, 6, 18

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Manuel Y . Galliker, Dibya Ghosh, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Devin LeBlanc, Sergey Levine, Adrian Li-Bell, Mo- hith Mothukuri, Suraj Nair, Karl Pertsch, Allen Z. Ren, Lucy...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    RT-1: Robotics transformer for real-world control at scale

    Anthony Brohan, Noah Brown, Justice Carbajal, Yev- gen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Jasmine Hsu, et al. RT-1: Robotics transformer for real-world control at scale. InRSS, 2023. 1

    work page 2023

  10. [10]

    Genie: Generative interactive environments

    Jake Bruce, Michael D Dennis, Ashley Edwards, Jack Parker-Holder, Yuge Shi, Edward Hughes, Matthew Lai, Aditi Mavalankar, Richie Steigerwald, Chris Apps, et al. Genie: Generative interactive environments. InICML,

    work page

  11. [11]

    AgiBot World Colosseo: A large-scale manipulation platform for scalable and intelligent embodied systems

    Qingwen Bu, Jisong Cai, Li Chen, Xiuqi Cui, Yan Ding, Siyuan Feng, Xindong He, Xu Huang, et al. AgiBot World Colosseo: A large-scale manipulation platform for scalable and intelligent embodied systems. InIROS,

    work page

  12. [12]

    Univla: Learning to act anywhere with task-centric latent actions

    Qingwen Bu, Yanting Yang, Jisong Cai, Shenyuan Gao, Guanghui Ren, Maoqing Yao, Ping Luo, and Hongyang Li. Univla: Learning to act anywhere with task-centric latent actions. InRSS, 2025. 18

    work page 2025

  13. [13]

    Diwa: Diffusion policy adaptation with world models.arXiv preprint arXiv:2508.03645,

    Akshay L Chandra, Iman Nematollahi, Chenguang Huang, Tim Welschehold, Wolfram Burgard, and Ab- hinav Valada. DiW A: Diffusion policy adaptation with world models.arXiv preprint arXiv:2508.03645, 2025. 5, 8

    work page arXiv 2025

  14. [14]

    Kang Chen, Zhihao Liu, Tonghe Zhang, Zhen Guo, Si Xu, Hao Lin, Hongzhi Zang, Quanlu Zhang, Zhaofei Yu, Guoliang Fan, et al.π RL: Online rl fine-tuning for flow-based vision-language-action models.arXiv preprint arXiv:2510.25889, 2025. 8

    work page arXiv 2025

  15. [15]

    Intelli- gent robot manipulation requires self-directed learning

    Li Chen, Chonghao Sima, Kashyap Chitta, Antonio Loquercio, Ping Luo, Yi Ma, and Hongyang Li. Intelli- gent robot manipulation requires self-directed learning. OpenReview, 2026. URL https://openreview.net/forum? id=Seb7rprW1Y. Accessed: 2026-01-02. 2

    work page 2026

  16. [16]

    RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation

    Tianxing Chen, Zanxin Chen, Baijun Chen, Zijian Cai, Yibin Liu, Zixuan Li, Qiwei Liang, Xianliang Lin, Yiheng Ge, Zhenyu Gu, et al. RoboTwin 2.0: A scalable data generator and benchmark with strong domain ran- domization for robust bimanual robotic manipulation. arXiv preprint arXiv:2506.18088, 2025. 8

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  17. [17]

    arXiv preprint arXiv:2506.08440 , year=

    Zengjue Chen, Runliang Niu, He Kong, Qi Wang, Qianli Xing, and Zipei Fan. TGRPO: Fine- tuning vision-language-action model via trajectory-wise group relative policy optimization.arXiv preprint arXiv:2506.08440, 2025. 8

    work page arXiv 2025

  18. [18]

    Dif- fusion Policy: Visuomotor policy learning via action diffusion

    Cheng Chi, Siyuan Feng, Yilun Du, Zhenjia Xu, Eric Cousineau, Benjamin Burchfiel, and Shuran Song. Dif- fusion Policy: Visuomotor policy learning via action diffusion. InRSS, 2023. 18

    work page 2023

  19. [19]

    Universal Manipulation Interface: In-the- wild robot teaching without in-the-wild robots

    Cheng Chi, Zhenjia Xu, Chuer Pan, Eric Cousineau, Benjamin Burchfiel, Siyuan Feng, Russ Tedrake, and Shuran Song. Universal Manipulation Interface: In-the- wild robot teaching without in-the-wild robots. InRSS,

    work page

  20. [20]

    Tenenbaum, Leslie Pack Kaelbling, Andy Zeng, and Jonathan Tompson

    Yilun Du, Sherry Yang, Pete Florence, Fei Xia, Ayzaan Wahid, Brian Ichter, Pierre Sermanet, Tianhe Yu, Pieter Abbeel, Joshua B. Tenenbaum, Leslie Pack Kaelbling, Andy Zeng, and Jonathan Tompson. Video Language Planning. InICLR, 2024. 2, 8

    work page 2024

  21. [21]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. InForty-first international conference on machine learning, 2024. 15

    work page 2024

  22. [22]

    MOKA: Open-World Robotic Manipulation through Mark-Based Visual Prompting

    Kuan Fang, Fangchen Liu, Pieter Abbeel, and Sergey Levine. MOKA: Open-World Robotic Manipulation through Mark-Based Visual Prompting. InRSS, 2024. 1

    work page 2024

  23. [23]

    Diffusion guidance is a controllable policy im- provement operator.arXiv preprint arXiv:2505.23458,

    Kevin Frans, Seohong Park, Pieter Abbeel, and Sergey Levine. Diffusion guidance is a controllable policy im- provement operator.arXiv preprint arXiv:2505.23458,

    work page arXiv

  24. [24]

    Adaworld: Learning adaptable world models with latent actions.arXiv preprint arXiv:2503.18938, 2025

    Shenyuan Gao, Siyuan Zhou, Yilun Du, Jun Zhang, and Chuang Gan. AdaWorld: Learning adaptable world models with latent actions.arXiv preprint arXiv:2503.18938, 2025. 8

    work page arXiv 2025

  25. [25]

    Self-improving embodied foundation models

    Seyed Kamyar Seyed Ghasemipour, Ayzaan Wahid, Jonathan Tompson, Pannag Sanketi, and Igor Mordatch. Self-improving embodied foundation models.arXiv preprint arXiv:2509.15155, 2025. 2, 8, 18

    work page arXiv 2025

  26. [26]

    Ctrl-World: A Controllable Generative World Model for Robot Manipulation

    Yanjiang Guo, Lucy Xiaoyang Shi, Jianyu Chen, and Chelsea Finn. Ctrl-world: A controllable generative world model for robot manipulation.arXiv preprint arXiv:2510.10125, 2025. 2, 8

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [27]

    Recurrent World Models Facilitate Policy Evolution

    David Ha and J ¨urgen Schmidhuber. Recurrent World Models Facilitate Policy Evolution. InNeurIPS, 2018. 2, 8

    work page 2018

  28. [28]

    LTX-Video: Realtime Video Latent Diffusion

    Yoav HaCohen, Nisan Chiprut, Benny Brazowski, Daniel Shalem, Dudu Moshe, Eitan Richardson, Eran Levin, Guy Shiran, Nir Zabari, Ori Gordon, et al. LTX- Video: Realtime video latent diffusion.arXiv preprint arXiv:2501.00103, 2024. 2, 4

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  29. [29]

    Dream to Control: Learning Behaviors by Latent Imagination

    Danijar Hafner, Timothy Lillicrap, Jimmy Ba, and Mohammad Norouzi. Dream to Control: Learning Behaviors by Latent Imagination.arXiv preprint arXiv:1912.01603, 2019. 2, 8

    work page internal anchor Pith review Pith/arXiv arXiv 1912

  30. [30]

    Mastering Atari with Discrete World Models

    Danijar Hafner, Timothy Lillicrap, Mohammad Norouzi, and Jimmy Ba. Mastering Atari with Discrete World Models. InICLR, 2021. 2, 8

    work page 2021

  31. [31]

    Mastering Diverse Domains through World Models

    Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Tim- othy Lillicrap. Mastering Diverse Domains through World Models.arXiv preprint arXiv:2301.04104, 2023. 2, 8

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  32. [32]

    Training Agents Inside of Scalable World Models

    Danijar Hafner, Wilson Yan, and Timothy Lillicrap. Training agents inside of scalable world models.arXiv preprint arXiv:2509.24527, 2025. 8

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  33. [33]

    Philippe Hansen-Estruch, Ilya Kostrikov, Michael Janner, Jakub Grudzien Kuba, and Sergey Levine

    Nicklas Hansen, Yixin Lin, Hao Su, Xiaolong Wang, Vikash Kumar, and Aravind Rajeswaran. MoDem: Accelerating visual model-based reinforcement learning with demonstrations.arXiv preprint arXiv:2212.05698,

    work page arXiv

  34. [34]

    Temporal difference learning for model predictive control

    Nicklas Hansen, Hao Su, and Xiaolong Wang. Temporal difference learning for model predictive control. In ICML, 2022. 8

    work page 2022

  35. [35]

    TD-MPC2: Scalable, Robust World Models for Continuous Control

    Nicklas Hansen, Hao Su, and Xiaolong Wang. TD- MPC2: Scalable, robust world models for continuous control.arXiv preprint arXiv:2310.16828, 2023. 8

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  36. [36]

    Hierarchical world models as visual whole-body humanoid controllers.arXiv preprint arXiv:2405.18418, 2024

    Nicklas Hansen, Jyothir SV , Vlad Sobal, Yann LeCun, Xiaolong Wang, and Hao Su. Hierarchical world models as visual whole-body humanoid controllers.arXiv preprint arXiv:2405.18418, 2024. 8

    work page arXiv 2024

  37. [37]

    Rac: Robot learning for long-horizon tasks by scaling recovery and correction.arXiv preprint arXiv:2509.07953, 2025

    Zheyuan Hu, Robyn Wu, Naveen Enock, Jasmine Li, Riya Kadakia, Zackory Erickson, and Aviral Ku- mar. RaC: Robot learning for long-horizon tasks by scaling recovery and correction.arXiv preprint arXiv:2509.07953, 2025. 2

    work page arXiv 2025

  38. [38]

    Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

    Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self Forcing: Bridging the train-test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    NORA-1.5: A vision- language-action model trained using world model-, and action-based preference rewards,

    Chia-Yu Hung, Navonil Majumder, Haoyuan Deng, Liu Renhang, Yankang Ang, Amir Zadeh, Chuan Li, Dorien Herremans, Ziwei Wang, and Soujanya Poria. NORA-1.5: A vision-language-action model trained us- ing world model-and action-based preference rewards. arXiv preprint arXiv:2511.14659, 2025. 8

    work page arXiv 2025

  40. [40]

    Vetrov, and Andrew Gordon Wilson

    Pavel Izmailov, Dmitrii Podoprikhin, Timur Garipov, Dmitry P. Vetrov, and Andrew Gordon Wilson. Av- eraging weights leads to wider optima and better gen- eralization. InUAI, 2018. 5

    work page 2018

  41. [41]

    DreamGen: Un- locking generalization in robot learning through video world models

    Joel Jang, Seonghyeon Ye, Zongyu Lin, Jiannan Xiang, Johan Bjorck, Yu Fang, Fengyuan Hu, Spencer Huang, Kaushil Kundalia, Yen-Chen Lin, et al. DreamGen: Un- locking generalization in robot learning through video world models. InCoRL, 2025. 8, 18

    work page 2025

  42. [42]

    Whole- BodyVLA: Towards unified latent vla for whole-body loco-manipulation control

    Haoran Jiang, Jin Chen, Qingwen Bu, Li Chen, Modi Shi, Yanjie Zhang, Delong Li, Chuanzhe Suo, Chuang Wang, Zhihui Peng, and Hongyang Li. Whole- BodyVLA: Towards unified latent vla for whole-body loco-manipulation control. InICLR, 2026. 18

    work page 2026

  43. [43]

    Galaxea open-world dataset and G0 dual-system VLA model.arXiv preprint arXiv:2509.00576, 2025

    Tao Jiang, Tianyuan Yuan, Yicheng Liu, Chenhao Lu, Jianning Cui, Xiao Liu, Shuiqi Cheng, Jiyang Gao, Huazhe Xu, and Hang Zhao. Galaxea open-world dataset and g0 dual-system vla model.arXiv preprint arXiv:2509.00576, 2025. 4, 6, 18

    work page arXiv 2025

  44. [44]

    World4rl: Diffusion world models for policy refinement with reinforcement learning for robotic manipulation,

    Zhennan Jiang, Kai Liu, Yuxin Qin, Shuai Tian, Yu- peng Zheng, Mingcai Zhou, Chao Yu, Haoran Li, and Dongbin Zhao. World4RL: Diffusion world models for policy refinement with reinforcement learning for robotic manipulation.arXiv preprint arXiv:2509.19080,

    work page arXiv

  45. [45]

    HG-DAgger: Interactive imitation learning with human experts

    Michael Kelly, Chelsea Sidrane, Katherine Driggs- Campbell, and Mykel J Kochenderfer. HG-DAgger: Interactive imitation learning with human experts. In ICRA, 2019. 2, 6

    work page 2019

  46. [46]

    Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Paul Foster, Pannag R. Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. OpenVLA: An open-source vision-language- action model. InCoRL, 2024. 1, 18

    work page 2024

  47. [47]

    Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success

    Moo Jin Kim, Chelsea Finn, and Percy Liang. Fine-tuning vision-language-action models: Optimizing speed and success.arXiv preprint arXiv:2502.19645,

    work page internal anchor Pith review Pith/arXiv arXiv

  48. [48]

    Reward-conditioned policies

    Aviral Kumar, Xue Bin Peng, and Sergey Levine. Reward-conditioned policies.arXiv preprint arXiv:1912.13465, 2019. 6, 8

    work page arXiv 1912

  49. [49]

    MoDem-V2: Visuo-motor world models for real-world robot manipulation

    Patrick Lancaster, Nicklas Hansen, Aravind Rajeswaran, and Vikash Kumar. MoDem-V2: Visuo-motor world models for real-world robot manipulation. InICRA,

    work page

  50. [50]

    A path towards autonomous machine intelligence.Open Review, 2022

    Yann LeCun. A path towards autonomous machine intelligence.Open Review, 2022. 2, 8

    work page 2022

  51. [51]

    Rl-100: Performant robotic manipulation with real-world reinforcement learning, 2025

    Kun Lei, Huanyu Li, Dongjie Yu, Zhenyu Wei, Lingxiao Guo, Zhennan Jiang, Ziyu Wang, Shiyu Liang, and Huazhe Xu. RL-100: Performant robotic manipulation with real-world reinforcement learning.arXiv preprint arXiv:2510.14830, 2025. 8

    work page arXiv 2025

  52. [52]

    Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems

    Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial, review, and perspectives on open problems.arXiv preprint arXiv:2005.01643, 2020. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2005

  53. [53]

    BEHAVIOR-1K: A Human-Centered, Embodied AI Benchmark with 1,000 Everyday Activities and Realistic Simulation

    Chengshu Li, Ruohan Zhang, Josiah Wong, Cem Gok- men, Sanjana Srivastava, Roberto Mart ´ın-Mart´ın, Chen Wang, Gabrael Levine, Wensi Ai, Benjamin Martinez, et al. BEHA VIOR-1K: A human-centered, embodied ai benchmark with 1,000 everyday activities and realistic simulation.arXiv preprint arXiv:2403.09227, 2024. 18

    work page internal anchor Pith review arXiv 2024

  54. [54]

    Robotic world model: A neural network simulator for robust policy optimization in robotics.arXiv preprint arXiv:2501.10100, 2025a

    Chenhao Li, Andreas Krause, and Marco Hutter. Robotic World Model: A neural network simulator for robust policy optimization in robotics.arXiv preprint arXiv:2501.10100, 2025. 8

    work page arXiv 2025

  55. [55]

    Li, S., Wu, K., Zhang, C., and Zhu, Y

    Dacheng Li, Yunhao Fang, Yukang Chen, Shuo Yang, Shiyi Cao, Justin Wong, Michael Luo, Xiaolong Wang, Hongxu Yin, Joseph E Gonzalez, et al. WorldMod- elBench: Judging video generation models as world models.arXiv preprint arXiv:2502.20694, 2025. 2

    work page arXiv 2025

  56. [56]

    SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning

    Haozhan Li, Yuxin Zuo, Jiale Yu, Yuhao Zhang, Zhao- hui Yang, Kaiyan Zhang, Xuekai Zhu, Yuchen Zhang, Tianxing Chen, Ganqu Cui, et al. SimpleVLA-RL: Scaling vla training via reinforcement learning.arXiv preprint arXiv:2509.09674, 2025. 2, 8

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  57. [57]

    A comprehensive survey on world models for embodied ai.arXiv preprint arXiv:2510.16732, 2025

    Xinqing Li, Xin He, Le Zhang, Min Wu, Xiaoli Li, and Yun Liu. A comprehensive survey on world models for embodied ai.arXiv preprint arXiv:2510.16732, 2025. 2

    work page arXiv 2025

  58. [58]

    Gr-rl: Going dexterous and precise for long-horizon robotic manipulation

    Yunfei Li, Xiao Ma, Jiafeng Xu, Yu Cui, Zhongren Cui, Zhigang Han, Liqun Huang, Tao Kong, Yuxiao Liu, Hao Niu, et al. GR-RL: Going dexterous and precise for long-horizon robotic manipulation.arXiv preprint arXiv:2512.01801, 2025. 8

    work page arXiv 2025

  59. [59]

    Genie Envisioner: A Unified World Foundation Platform for Robotic Manipulation

    Yue Liao, Pengfei Zhou, Siyuan Huang, Donglin Yang, Shengcong Chen, Yuxin Jiang, Yue Hu, Jingbin Cai, Si Liu, Jianlan Luo, et al. Genie Envisioner: A unified world foundation platform for robotic manipulation. arXiv preprint arXiv:2508.05635, 2025. 2, 4, 7, 8, 17, 18

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  60. [60]

    LIBERO: Benchmark- ing knowledge transfer for lifelong robot learning

    Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. LIBERO: Benchmark- ing knowledge transfer for lifelong robot learning. In NeurIPS, 2023. 2, 8

    work page 2023

  61. [61]

    What can rl bring to vla generalization? an empirical study

    Jijia Liu, Feng Gao, Bingwen Wei, Xinlei Chen, Qing- min Liao, Yi Wu, Chao Yu, and Yu Wang. What can rl bring to vla generalization? an empirical study. In NeurIPS, 2025. 2, 8

    work page 2025

  62. [62]

    RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation

    Songming Liu, Lingxuan Wu, Bangguo Li, Hengkai Tan, Huayu Chen, Zhengyi Wang, Ke Xu, Hang Su, and Jun Zhu. RDT-1B: A diffusion foundation model for bi- manual manipulation.arXiv preprint arXiv:2410.07864,

    work page internal anchor Pith review Pith/arXiv arXiv

  63. [63]

    VLA-RL: Towards Masterful and General Robotic Manipulation with Scalable Reinforcement Learning

    Guanxing Lu, Wenkai Guo, Chubin Zhang, Yuheng Zhou, Haonan Jiang, Zifeng Gao, Yansong Tang, and Ziwei Wang. VLA-RL: Towards masterful and general robotic manipulation with scalable reinforcement learn- ing.arXiv preprint arXiv:2505.18719, 2025. 2, 8

    work page internal anchor Pith review arXiv 2025

  64. [64]

    SERL: A software suite for sample-efficient robotic reinforcement learning

    Jianlan Luo, Zheyuan Hu, Charles Xu, You Liang Tan, Jacob Berg, Archit Sharma, Stefan Schaal, Chelsea Finn, Abhishek Gupta, and Sergey Levine. SERL: A software suite for sample-efficient robotic reinforcement learning. InICRA, 2024. 2, 8

    work page 2024

  65. [65]

    Precise and dexterous robotic manipulation via human-in-the-loop reinforcement learning.Science Robotics, 2025

    Jianlan Luo, Charles Xu, Jeffrey Wu, and Sergey Levine. Precise and dexterous robotic manipulation via human-in-the-loop reinforcement learning.Science Robotics, 2025. 2, 8, 17

    work page 2025

  66. [66]

    Vision language models are in-context value learners

    Yecheng Jason Ma, Joey Hejna, Chuyuan Fu, Dhruv Shah, Jacky Liang, Zhuo Xu, Sean Kirmani, Peng Xu, Danny Driess, Ted Xiao, et al. Vision language models are in-context value learners. InICLR, 2024. 2, 8

    work page 2024

  67. [67]

    CALVIN: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks.RA-L, 2022

    Oier Mees, Lukas Hermann, Erick Rosete-Beas, and Wolfram Burgard. CALVIN: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks.RA-L, 2022. 8

    work page 2022

  68. [68]

    Structured world models from human videos

    Russell Mendonca, Shikhar Bahl, and Deepak Pathak. Structured world models from human videos. InCoRL,

    work page

  69. [69]

    RoboTwin: Dual-arm robot benchmark with generative digital twins

    Yao Mu, Tianxing Chen, Zanxin Chen, Shijia Peng, Zhiqian Lan, Zeyu Gao, Zhixuan Liang, Qiaojun Yu, Yude Zou, Mingkun Xu, et al. RoboTwin: Dual-arm robot benchmark with generative digital twins. In CVPR, 2025. 8

    work page 2025

  70. [70]

    RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots

    Soroush Nasiriany, Abhiram Maddukuri, Lance Zhang, Adeet Parikh, Aaron Lo, Abhishek Joshi, Ajay Man- dlekar, and Yuke Zhu. RoboCasa: Large-scale simu- lation of everyday tasks for generalist robots.arXiv preprint arXiv:2406.02523, 2024. 18

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  71. [71]

    Open X-Embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collabo- ration

    Abby O’Neill, Abdul Rehman, Abhiram Maddukuri, Abhishek Gupta, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, Ajinkya Jain, et al. Open X-Embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collabo- ration. InICRA, 2024. 18

    work page 2024

  72. [72]

    Advantage-Weighted Regression: Simple and Scalable Off-Policy Reinforcement Learning

    Xue Bin Peng, Aviral Kumar, Grace Zhang, and Sergey Levine. Advantage-weighted regression: Simple and scalable off-policy reinforcement learning.arXiv preprint arXiv:1910.00177, 2019. 2

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  73. [73]

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

    Stephane Ross, Geoffrey Gordon, and Drew Bagnell. A reduction of imitation learning and structured prediction to no-regret online learning. InAISTATS, 2011. 2, 6

    work page 2011

  74. [74]

    Learned perceptive forward dynamics model for safe and platform-aware robotic navigation

    Pascal Roth, Jonas Frey, Cesar Cadena, and Marco Hutter. Learned perceptive forward dynamics model for safe and platform-aware robotic navigation. InRSS,

    work page

  75. [75]

    Proximal Policy Optimization Algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy opti- mization algorithms.arXiv preprint arXiv:1707.06347,

    work page internal anchor Pith review Pith/arXiv arXiv

  76. [76]

    Is Diversity All You Need for Scalable Robotic Manipulation?

    Modi Shi, Li Chen, Jin Chen, Yuxiang Lu, Chiming Liu, Guanghui Ren, Ping Luo, Di Huang, Maoqing Yao, and Hongyang Li. Is diversity all you need for scalable robotic manipulation?arXiv preprint arXiv:2507.06219,

    work page arXiv

  77. [77]

    Richard S. Sutton. Learning to predict by the methods of temporal differences.Machine learning, 1988. 2, 5, 8

    work page 1988

  78. [78]

    Richard S. Sutton. Dyna, an integrated architecture for learning, planning, and reacting.ACM Sigart Bulletin,

    work page

  79. [79]

    Fvd: A new metric for video generation

    Thomas Unterthiner, Sjoerd Van Steenkiste, Karol Ku- rach, Rapha ¨el Marinier, Marcin Michalski, and Sylvain Gelly. Fvd: A new metric for video generation. 2019. 8

    work page 2019

  80. [80]

    Steer- ing your diffusion policy with latent space reinforce- ment learning

    Andrew Wagenmaker, Yunchu Zhang, Mitsuhiko Nakamoto, Seohong Park, Waleed Yagoub, Anusha Nagabandi, Abhishek Gupta, and Sergey Levine. Steer- ing your diffusion policy with latent space reinforce- ment learning. InCoRL, 2025. 2, 6, 8, 14, 16, 17

    work page 2025

Showing first 80 references.