Pith. sign in

REVIEW 2 major objections 5 minor 34 references

ModPack claims a single wearable backpack can teleoperate diverse bimanual mobile robots through plug-and-play modules.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 12:36 UTC pith:ZLHAYOEZ

load-bearing objection Genuinely modular teleoperation system; haptic and active-perception claims outrun the evidence, but the engineering is solid and the paper deserves a serious referee. the 2 major comments →

arxiv 2607.19479 v1 pith:ZLHAYOEZ submitted 2026-07-21 cs.RO cs.AI

ModPack: An Extensible Teleoperation Interface for Bimanual Mobile Manipulation

Joshua Citron , Renee Zbizika , Zeyi Liu , Shuran Song This is my paper
classification cs.RO cs.AI
keywords teleoperationmobile manipulationimitation learningmodular hardwarehaptic feedbackactive perceptionwearable interfacedata collection
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper argues that teleoperation does not have to be rebuilt from scratch for every robot or task. It introduces ModPack, a wearable backpack that carries compute, power, communication, and storage, with plug-and-play modules for joint-level arm control, mobile-base control, active perception, and haptic feedback. The authors claim this modular design lets one interface handle two different robot platforms and two real-world mobile manipulation tasks, and that policies trained on demonstrations collected through ModPack achieve strong deployment performance. If true, robot-learning groups could reuse a single teleoperation interface across embodiments, swapping only the modules a task needs.

Core claim

The central claim is that a self-contained backpack substrate can decouple shared teleoperation infrastructure from task-specific and robot-specific capabilities, and that this decoupling is enough to build a reusable data-collection and policy-learning platform. ModPack supports cross-robot joint-level control through leader arms kinematically matched to each follower arm, mobile base control by tracking the operator's body motion, active perception by streaming the robot's egocentric view to a VR headset while compensating for base motion, and haptic feedback by mapping end-effector forces back to the leader arms via a Jacobian transpose. The authors validate the system by collecting 125 d

What carries the argument

The load-bearing mechanism is the modular backpack core plus its software bridge: a lightweight message-queue API that lets each robot expose its own configuration while ModPack orchestrates modules and logs data. On top of this substrate sit the leader-arm modules—dynamically actuated arms designed to be kinematically equivalent to their follower arms—with active gravity compensation to reduce operator fatigue and, when force sensors are present, haptic feedback computed as the Jacobian transpose of the measured end-effector wrench. The mobile-base module recovers the operator's true center of rotation from a backpack-mounted iPhone, and the active-perception module maps head pose to neck c

Load-bearing premise

The claim that haptic feedback during teleoperation makes demonstrations safer and more precise is asserted but never directly tested, so if haptics add little to demonstration quality, a key advantage claimed for the system is unsupported.

What would settle it

Collect Box Transfer demonstrations with haptic feedback disabled, using the same operator and task setup, train the same vision-only policy on those demos, and compare deployment success rates against the haptic-enabled set; if success does not drop, the haptic benefit claim is falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the modular abstraction holds, adapting ModPack to a new robot arm mainly requires designing a new kinematically equivalent leader arm, not redesigning the entire teleoperation stack.
  • Operators can simultaneously command arms, base, and head with motion disentangling, enabling demonstrations for long-horizon tasks that need search, navigation, and manipulation.
  • Haptic feedback with current budgeting could make contact-rich data collection feasible on wearable, untethered systems without heavy stationary force-feedback hardware.
  • Open-sourcing the hardware and software means other labs can reproduce the system and add their own modules through the same bridge API, potentially standardizing demonstration collection across labs.
  • The reported policy success rates suggest that data collected through a modular teleoperation interface is usable for imitation learning on real mobile manipulation tasks.

Where Pith is reading between the lines

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

  • If the modular interface is as clean as claimed, it could serve as a common testbed for studying cross-embodiment policy transfer, since the same demonstration format would span arms with different degrees of freedom—though that would require designing new leader arms for each robot, which may limit practical speed.
  • The haptic-feedback benefit is not isolated by the experiments: the ablation removes torque as a policy input, not haptic feedback during data collection. A fair test would collect demos with haptics disabled and compare downstream policy success.
  • The head/base decoupling in the active-perception module could be reused to explicitly study viewpoint selection as a learnable behavior, since the policy attention analysis hints that torque tokens matter most during contact-critical phases.
  • The stated motor-torque and battery limits suggest that scaling to heavier manipulation tasks may hit wearable hardware constraints before algorithmic ones, so future gains may depend on lighter, higher-torque actuators.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. ModPack is a modular teleoperation system built around a wearable backpack providing onboard compute, power, communication, and data logging. The paper describes plug-and-play modules for joint-level leader-arm teleoperation with gravity compensation and haptic feedback, mobile-base control via an iPhone/WebXR session, and active perception via an Apple Vision Pro headset. The system is evaluated on two different bimanual mobile robots: a customized ARX5 platform for a cloth placement task with active perception (125 demonstrations, 25 rollouts per policy variant) and an RB-Y1m robot for a box transfer task with haptic feedback (102 demonstrations, 20 rollouts per policy variant). Policies are trained with a transformer-based Diffusion Policy. The paper's central claim is that ModPack is a flexible, reusable data-collection interface, and that policies trained on ModPack-collected data achieve strong deployment performance. The authors state they will open-source the hardware design and software stack.

Significance. If the system performs as claimed, ModPack addresses a real bottleneck in robot learning: the fragmentation of teleoperation interfaces across robot embodiments and tasks. The paper's strengths include a complete and detailed hardware/software design, a genuinely modular architecture, deployment on two distinct real robot platforms, and open-sourcing of the full system. These are substantial practical contributions for the imitation-learning community. However, the headline evaluative claims—especially the value of the haptic-feedback module and the overall 'strong deployment performance'—rest on experiments with limited statistical power and one confounded comparison. The modularity and extensibility of the system are credible from the engineering and the two deployments; the specific performance claims need stronger evidence or more conservative wording.

major comments (2)
  1. [§3.2.1, §4.2, Fig. 5(c)] The claim that haptic feedback enables 'safer and more precise teleoperation' (Sec. 3.2.1) and that the Box Transfer experiments validate ModPack's utility as a data-collection interface (Sec. 6) is not supported by the reported experiment. All 102 Box Transfer demonstrations were collected with haptic feedback active (Sec. 4.2). The comparison [All Cam] vs [All Cam + Torque] ablates whether joint torques are provided as observations to the policy at training/inference time; it does not vary whether the operator received haptic feedback during data collection. The 12/20 vs 6/20 gap could therefore be due entirely to the torque conditioning signal, not to improved demonstration quality. To support the haptic-feedback claim, the paper needs either (a) a comparison in which demonstrations are collected with haptic feedback disabled while all other conditions are held fixed, or (b) direct me
  2. [§4.1, §4.2, Figs. 4(c), 5(c)] Success rates are reported as raw counts from single evaluation runs (25 or 20 rollouts per policy) with no confidence intervals, no repeated seeds, and no statistical tests. Several key comparisons are within the noise range for these sample sizes: 22/25 vs 20/25 in Cloth Placement, and 12/20 vs 11/20 in Box Transfer ([All Cam + Torque] vs [Head Cam]). The conclusion that policies 'achieve strong deployment performance' (Sec. 6) therefore goes beyond the evidence reported. Please provide per-seed results, confidence intervals, or at minimum a paired significance test. If that is not feasible, soften the strong language and present the results as pilot validation of the data-collection pipeline rather than as a ranked comparison of policy variants.
minor comments (5)
  1. [§4 (opening)] Typo: 'adaptibility' should be 'adaptability'.
  2. [Appendix F.1] Typo: 'RBG images' should be 'RGB images'.
  3. [Fig. 5 caption] The caption says the policies are compared against 'head-camera-only and wrist-camera-only ablations,' but the figure actually includes [All Cam] and [All Cam + Torque]; the caption should be updated to describe all four variants.
  4. [§4.2, Fig. 6] The attention analysis is qualitative and based on a single or few rollouts. The statement that 'attention weights increase' during contact-critical phases would benefit from variance across rollouts or a quantitative measure; otherwise it should be described as an illustrative observation.
  5. [Appendix D.2.2] The haptic gain k_c, rate limit, and left-arm scale are mentioned in Sec. 3.2.1 and Appendix D.2.2, but k_c is never given a value or a tuning procedure. Please provide the actual parameter range used in experiments for reproducibility.

Circularity Check

0 steps flagged

No significant circularity: ModPack's claims are validated by real deployment rollouts, and its kinematic/haptic equations are standard mappings rather than self-referential fits.

full rationale

This is an empirical systems paper, not a derivation that reduces to its own outputs. The central claim—that policies trained on ModPack-collected data achieve strong deployment performance—is supported by actual policy rollouts (e.g., 22/25, 12/20 success rates), which are independent of the training demonstrations and are not refits of fitted parameters. The haptic feedback law τ_c = k_c J^T F_ee is a standard Jacobian-transpose force mapping, and the base/head transforms are ordinary coordinate-frame kinematics; none of these equations are defined in terms of the downstream success metrics. The paper does reuse the authors' prior systems for components such as the holonomic base [12], active perception [28], and message-queue communication [30], but these are implementation choices and not invoked as load-bearing evidence for ModPack's utility. The one legitimate weakness—that the haptic feedback benefit is asserted from a comparison that ablates torque as a policy input rather than haptic feedback during data collection—is an experimental confound and a correctness-risk concern, not a circular reduction. Accordingly, the appropriate circularity score is 0.

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 0 invented entities

No new physical or theoretical entities are introduced. The system's empirical performance depends on several hand-tuned control parameters and calibration offsets, which are listed as free parameters. The axioms are standard robotics modeling assumptions inherited from the design choices (GELLO kinematics, KDL dynamics, WebXR tracking, Jacobian transpose haptics).

free parameters (7)
  • Damping gain matrix K_d = 0.1 per joint (Appendix D.1.2)
    Empirically chosen to reduce jitter in gravity compensation; affects leader-arm stability and haptic transparency.
  • EMA smoothing factor alpha = not reported
    Used in Equation (1) to filter gravity-compensation torques; value is hand-tuned for stability.
  • Haptic gain k_c = not reported
    Scales Jacobian-transpose force feedback (Section 3.2.1); tuneable and varies with the robot and task.
  • Haptic rate limit and left-arm scale = 50 Nm/s, scale 0.8 (Appendix D.2.2)
    Empirically set to prevent jitter on the RB-Y1m leader arm; affects haptic feedback quality.
  • Joint current limits = e.g., joint 1: 100 mA (6-DoF), 200 mA (7-DoF) etc.
    Empirically determined per joint to provide stable gravity compensation without excessive stiffness; listed in Appendices D.1.2 and D.2.2.
  • WebXR device offset p_d = fixed for iPhone 17, must be re-measured for other models (Appendix E.1)
    Calibration offset between iPhone frame and camera mount center; affects base-motion tracking accuracy.
  • Operator center-of-rotation offset p_u = measured from head center to mount center (Appendix E.1)
    Operator-specific offset used to recover the user's true center of rotation; affects base control during walking.
axioms (4)
  • domain assumption Leader arms are kinematically equivalent to follower arms (GELLO [10])
    The paper relies on this to achieve direct joint-space teleoperation; if the replication is imperfect, control fidelity degrades (Section 3.2.1).
  • domain assumption Gravity compensation via KDL on the URDF model accurately represents leader-arm dynamics
    Used to compute G(q) for gravity compensation (Section 3.2.1); requires accurate mass/inertia parameters.
  • domain assumption WebXR/iPhone pose tracking is sufficiently accurate for egocentric base control
    The mobile-base module relies on iPhone WebXR pose estimates (Section 3.2.2); drift or latency would corrupt operator-to-robot mapping.
  • domain assumption Jacobian transpose mapping from end-effector wrench to joint torques provides adequate haptic feedback
    Haptic feedback uses tau_c = k_c * J^T F_ee (Section 3.2.1); the approximation ignores dynamics and may not faithfully render contact forces.

pith-pipeline@v1.3.0-alltime-deepseek · 12316 in / 10697 out tokens · 106900 ms · 2026-08-01T12:36:03.708895+00:00 · methodology

0 comments
read the original abstract

Existing teleoperation systems are often tailored to specific robot hardware and task domains, limiting their scalability and adaptability. We present ModPack, a modular and extensible teleoperation system designed to support diverse robot embodiments and task requirements within a unified framework. At the core of ModPack is a self-contained wearable "backpack" that integrates onboard computation, power, communication, and data storage. Built on top of this shared interface, the system supports plug-and-play capability modules including joint-level teleoperation with haptic feedback, mobile manipulation, and active perception. Experiments across two distinct robot platforms and real-world mobile manipulation tasks demonstrate that ModPack provides a flexible and reusable framework for data collection and policy learning. To support future research, we open-source the complete hardware design and software stack. Project website: https://modpack-robotics.github.io/

Figures

Figures reproduced from arXiv: 2607.19479 by Joshua Citron, Renee Zbizika, Shuran Song, Zeyi Liu.

Figure 1
Figure 1. Figure 1: ModPack is an extensible teleoperation interface for bimanual mobile manipulation, built around a wearable backpack base with plug-and-play capability modules. For example, the active perception module, shown on the left, lets operators control the robot’s head camera to search for cloth placement locations, while the leader arms module with haptic feedback, shown on the right, supports a box transfer task… view at source ↗
Figure 2
Figure 2. Figure 2: ModPack consists of a backpack core that supports plug-and-play modules, such as a Vision Pro for active perception and an iPhone for base control. Separate 6-DoF and 7-DoF leader arms are configured to control a customized mobile robot (top right) and an RB-Y1m robot (bottom right), respectively. TidyBot++ [12] supports joint control of a mobile base and a single arm, but does not scale naturally to coord… view at source ↗
Figure 3
Figure 3. Figure 3: Haptic feedback. The top plot shows the average joint torque on RB-Y1m left arm (blue), along with the corresponding haptic feedback torque norm commanded to the left leader arm (orange). The bottom panel shows operator’s response to the haptic feedback to adjust the force exerted on the box. Haptic Feedback To enhance operator sit￾uational awareness, we implement haptic feedback for follower robots equipp… view at source ↗
Figure 4
Figure 4. Figure 4: Summary of Cloth Placement task. a) The robot picks up a cloth, uses active perception to locate a basket placed on either side, navigates toward it, and places the cloth inside. b) Different evaluation scenarios with varying towel colors and basket locations. c) Task success rates across policy variants over input visual modalities. d) Failure cases across different policies. a) Policy Rollout Top b) Task… view at source ↗
Figure 5
Figure 5. Figure 5 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Attention Analysis. Results. We evaluate the task across two scenarios where either the top or bottom shelf is occupied and the robot has to place the box on the other unoccupied shelf. We compare three policy variants: one using all RGB cam￾era views (two head cameras and both wrist cameras) [All Cam], one using only the two head cameras [Head Cam], and one that takes in all camera views plus joint torque… view at source ↗
Figure 7
Figure 7. Figure 7: Circuit diagram for ModPack D Leader Arms D.1 ARX (6-dof) Leader Arm D.1.1 Motor Selection Motors were selected to provide the minimum amount of torque necessary to hold the arm for mass assumptions made using OnShape materials library. All motors are Dynamixel brand, with the models and control modes enumerated below. For this leader arm, we choose to not actuate joints 5 and 6 to allow for better wrist m… view at source ↗
Figure 8
Figure 8. Figure 8: Software diagram for ModPack. H Policy During training, we synchronize all sensory inputs to the downsampled head-camera timestamps (of￾ten 10-20 Hz). Proprioceptive states are interpolated at each head-camera timestamp, with positions and gripper states interpolated linearly and orientations interpolated using SLERP. Other modalities are aligned by selecting the nearest preceding measurement to the head-c… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 1 canonical work pages

  1. [1]

    Punamiya, S

    R. Punamiya, S. Kareer, Z. Liu, J. Citron, R.-Z. Qiu, X. Cai, A. Gavryushin, J. Chen, D. Li- conti, L. Y . Zhu, et al. Egoverse: An egocentric human dataset for robot learning from around the world.arXiv preprint arXiv:2604.07607, 2026

    Pith/arXiv arXiv 2026

  2. [2]

    Barreiros, A

    J. Barreiros, A. Beaulieu, A. Bhat, R. Cory, E. Cousineau, H. Dai, C.-H. Fang, K. Hashimoto, M. Z. Irshad, M. Itkina, et al. A careful examination of large behavior models for multitask dexterous manipulation.Science Robotics, 11(113):eaea6201, 2026

    2026

  3. [3]

    Bjorck, F

    J. Bjorck, F. Casta ˜neda, N. Cherniadev, X. Da, R. Ding, L. Fan, Y . Fang, D. Fox, F. Hu, S. Huang, et al. Gr00t n1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv:2503.14734, 2025

    Pith/arXiv arXiv 2025

  4. [4]

    G. R. Team, S. Abeyruwan, J. Ainslie, J.-B. Alayrac, M. G. Arenas, T. Armstrong, A. Balakr- ishna, R. Baruch, M. Bauza, M. Blokzijl, et al. Gemini robotics: Bringing ai into the physical world.arXiv preprint arXiv:2503.20020, 2025

    Pith/arXiv arXiv 2025

  5. [5]

    O’Neill, A

    A. O’Neill, A. Rehman, A. Maddukuri, A. Gupta, A. Padalkar, A. Lee, A. Pooley, A. Gupta, A. Mandlekar, A. Jain, et al. Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 6892–6903. IEEE, 2024

    2024

  6. [6]

    Khazatsky, K

    A. Khazatsky, K. Pertsch, S. Nair, A. Balakrishna, S. Dasari, S. Karamcheti, S. Nasiriany, M. K. Srirama, L. Y . Chen, K. Ellis, et al. Droid: A large-scale in-the-wild robot manipulation dataset.arXiv preprint arXiv:2403.12945, 2024

    Pith/arXiv arXiv 2024

  7. [7]

    B. D. Argall, S. Chernova, M. Veloso, and B. Browning. A survey of robot learning from demonstration.Robot. Auton. Syst., 57(5):469–483, May 2009. ISSN 0921-8890. doi:10. 1016/j.robot.2008.10.024. URLhttps://doi.org/10.1016/j.robot.2008.10.024

    doi:10.1016/j.robot.2008.10.024 2009

  8. [8]

    Shivin, A

    D. Shivin, A. Wensi, J. Yuqian, S. Samik, H. Jiaheng, Z. Ruohan, S. Peter, A. Ben, and M.-M. Roberto. Telemoma: A modular and versatile teleoperation system for mobile manipulation. arXiv (Cornell University), 2024. URLhttps://arxiv.org/abs/2403.07869

    Pith/arXiv arXiv 2024

  9. [9]

    A. Iyer, Z. Peng, Y . Dai, I. Guzey, S. Haldar, S. Chintala, and L. Pinto. Open teach: A versatile teleoperation system for robotic manipulation.arXiv preprint arXiv:2403.07870, 2024

    Pith/arXiv arXiv 2024

  10. [10]

    P. Wu, Y . Shentu, Z. Yi, X. Lin, and P. Abbeel. Gello: A general, low-cost, and intuitive teleoperation framework for robot manipulators. In2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 12156–12163. IEEE, 2024

    2024

  11. [11]

    Z. Fu, T. Z. Zhao, and C. Finn. Mobile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation.arXiv preprint arXiv:2401.02117, 2024

    Pith/arXiv arXiv 2024

  12. [12]

    J. Wu, W. Chong, R. Holmberg, A. Prasad, Y . Gao, O. Khatib, S. Song, S. Rusinkiewicz, and J. Bohg. Tidybot++: An open-source holonomic mobile manipulator for robot learning.arXiv preprint arXiv:2412.10447, 2024. 9

    Pith/arXiv arXiv 2024

  13. [13]

    Fang, H.-S

    H. Fang, H.-S. Fang, Y . Wang, J. Ren, J. Chen, R. Zhang, W. Wang, and C. Lu. Airexo: Low- cost exoskeletons for learning whole-arm manipulation in the wild. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 15031–15038. IEEE, 2024

    2024

  14. [14]

    H. Fang, C. Wang, Y . Wang, J. Chen, S. Xia, J. Lv, Z. He, X. Yi, Y . Guo, X. Zhan, et al. Airexo-2: Scaling up generalizable robotic imitation learning with low-cost exoskeletons. arXiv preprint arXiv:2503.03081, 2025

    Pith/arXiv arXiv 2025

  15. [15]

    Myers, O

    N. Myers, O. Kwon, S. Yamsani, and J. Kim. Child (controller for humanoid imitation and live demonstration): A whole-body humanoid teleoperation system. In2025 IEEE-RAS 24th International Conference on Humanoid Robots (Humanoids), pages 1–6. IEEE, 2025

    2025

  16. [16]

    O. Kwon, S. Yamsani, N. Myers, S. Taylor, J. Hong, K. Park, A. Alspach, and J. Kim. Paprle (plug-and-play robotic limb environment): A modular ecosystem for robotic limbs.arXiv preprint arXiv:2507.05555, 2025

    Pith/arXiv arXiv 2025

  17. [17]

    T. Chen, H. Wu, J. Wang, X. Li, Z. Jin, and L. Fang. Supersuit: An isomorphic bimodal interface for scalable mobile manipulation.arXiv preprint arXiv:2603.06280, 2026

    arXiv 2026

  18. [18]

    Lenz and S

    C. Lenz and S. Behnke. Bimanual telemanipulation with force and haptic feedback through an anthropomorphic avatar system.Robotics and Autonomous Systems, 161:104338, Mar. 2023. ISSN 0921-8890. doi:10.1016/j.robot.2022.104338. URLhttp://dx.doi.org/10.1016/ j.robot.2022.104338

    arXiv 2023

  19. [19]

    Schwarz, C

    M. Schwarz, C. Lenz, A. Rochow, M. Schreiber, and S. Behnke. Nimbro avatar: Interactive immersive telepresence with force-feedback telemanipulation. In2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 5312–5319, 2021. doi:10.1109/ IROS51168.2021.9636191

    arXiv 2021

  20. [20]

    Zhang and C

    Z. Zhang and C. Qian. Wearable teleoperation controller with 2-dof robotic arm and haptic feedback for enhanced interaction in virtual reality.Frontiers in Neu- rorobotics, V olume 17 - 2023, 2023. ISSN 1662-5218. doi:10.3389/fnbot. 2023.1228587. URLhttps://www.frontiersin.org/journals/neurorobotics/ articles/10.3389/fnbot.2023.1228587

    arXiv 2023

  21. [21]

    Purushottam, J

    A. Purushottam, J. Yan, C. Xu, Y . Sim, and J. Ramos. Wheeled humanoid bilateral tele- operation with position-force control modes for dynamic loco-manipulation, 2024. URL https://arxiv.org/abs/2407.12189

    Pith/arXiv arXiv 2024

  22. [22]

    F. Wu, Z. Jiao, W. Li, Z. Zhang, H. Li, J. Wu, B. Jia, and S. Dong. A vr-based robotic teleop- eration system with haptic feedback and adaptive collision avoidance.IEEE Transactions on Consumer Electronics, 71(3):8752–8761, 2025. doi:10.1109/TCE.2025.3580966

    arXiv 2025

  23. [23]

    R. Ding, Y . Qin, J. Zhu, C. Jia, S. Yang, R. Yang, X. Qi, and X. Wang. Bunny-visionpro: Real- time bimanual dexterous teleoperation for imitation learning. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 12248–12255. IEEE, 2025

    2025

  24. [24]

    X. Xu, J. Park, H. Zhang, E. Cousineau, A. Bhat, J. Barreiros, D. Wang, and S. Song. Hommi: Learning whole-body mobile manipulation from human demonstrations, 2026. URLhttps: //arxiv.org/abs/2603.03243

    Pith/arXiv arXiv 2026

  25. [25]

    Q. Zeng, C. Li, J. S. John, Z. Zhou, J. Wen, G. Feng, Y . Zhu, and Y . Xu. Activeumi: Robotic manipulation with active perception from robot-free human demonstrations, 2025. URLhttps://arxiv.org/abs/2510.01607

    arXiv 2025

  26. [26]

    J. Yu, Y . Shentu, D. Wu, P. Abbeel, K. Goldberg, and P. Wu. Egomi: Learning active vi- sion and whole-body manipulation from egocentric human demonstrations.arXiv preprint arXiv:2511.00153, 2025. 10

    arXiv 2025

  27. [27]

    Cheng, J

    X. Cheng, J. Li, S. Yang, G. Yang, and X. Wang. Open-television: Teleoperation with immer- sive active visual feedback.arXiv preprint arXiv:2407.01512, 2024

    Pith/arXiv arXiv 2024

  28. [28]

    Xiong, X

    H. Xiong, X. Xu, J. Wu, Y . Hou, J. Bohg, and S. Song. Vision in action: Learning active perception from human demonstrations. InConference on Robot Learning, pages 5450–5463. PMLR, 2025

    2025

  29. [29]

    Genitrini, F

    M. Genitrini, F. Dotti, E. Bianca, and A. Ferri. Impact of backpacks on ergonomics: Biome- chanical and physiological effects: A narrative review.International Journal of Environmental Research and Public Health, 19(11):6737, 2022. doi:10.3390/ijerph19116737

    work page doi:10.3390/ijerph19116737 2022

  30. [30]

    Y . Gao, J. Liu, S. Li, and S. Song. Gated memory policy, 2026. URLhttps://arxiv.org/ abs/2604.18933

    Pith/arXiv arXiv 2026

  31. [31]

    H. Shi, S. Hu, Y . Hou, W. Wang, K. Liu, and S. Song. Minimalist compliance control, 2026. URLhttps://arxiv.org/abs/2603.00913

    arXiv 2026

  32. [32]

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song. Diffusion policy: Visuomotor policy learning via action diffusion. the intl.Journal of Robotics Research, 2023

    2023

  33. [33]

    M. A. Lee, Y . Zhu, K. Srinivasan, P. Shah, S. Savarese, L. Fei-Fei, A. Garg, and J. Bohg. Making sense of vision and touch: Self-supervised learning of multimodal representations for contact-rich tasks. In2019 IEEE International Conference on Robotics and Automation (ICRA), 2019. URLhttps://arxiv.org/abs/1810.10191

    Pith/arXiv arXiv 2019

  34. [34]

    C. Lu, X. Cheng, J. Li, S. Yang, M. Ji, C. Yuan, G. Yang, S. Yi, and X. Wang. Mobile- television: Predictive motion priors for humanoid whole-body control. In2025 IEEE Interna- tional Conference on Robotics and Automation (ICRA), pages 5364–5371. IEEE, 2025. 11 Appendix A Comparison with Prior Works Cross-Robot Joint-Level Haptic Mobile Active Adaptabilit...

    2025