Pith. sign in

REVIEW 3 major objections 4 minor 43 references

UPTor: Unified 3D Human Pose Dynamics and Trajectory Prediction for Human-Robot Interaction

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read UPTor claims a single non-autoregressive transformer can predict full-body pose and global trajectory together in real time using a motion-transformation canonical frame.

desk verdict Useful coupling of pose and trajectory via a motion transformation and a new navigation dataset, but the evaluation leaks subjects in both navigation-focused splits. read the letter →

arxiv 2505.14866 v1 pith:LU6TTS3S submitted 2025-05-20 cs.RO cs.CV

Nisarga Nilavadi , Andrey Rudenko , Timm Linder This is my paper · ORCID
classification cs.ROcs.CV
keywords humanmotionpredictiontrajectoryfull-bodyposeforecastingnon-autoregressivetransformergraphattentionnetworkhuman-robotinteractionmobilerobotnavigationegocentricdataset
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

UPTor claims that full-body pose dynamics and the global trajectory of a walking person can be predicted by a single non-autoregressive transformer, provided the whole motion sequence is first normalized into a canonical coordinate frame. The normalization, called motion transformation, translates the last observed root position to the origin and rotates the observed motion direction onto the positive x-axis, so the model learns pose and root movement as one coupled prediction rather than two separate tasks. The paper reports that this compact model matches or improves on the decoupled STPOTR baseline on Human3.6M while using roughly half the transformer parameters, runs in 17 ms, and gives lower pose and trajectory errors on locomotion-heavy CMU-Mocap data and on the new DARKO robot-egocentric navigation dataset. A reader should care because real-time full-body forecasting in global coordinates is what a mobile robot needs to plan safe, socially aware navigation around people.

What carries the argument

The load-bearing object is the motion transformation: a per-sequence normalization that subtracts the root joint position at the last observed frame from every frame and then rotates all frames by the angle between the root's recent heading and the positive x-axis. This places the observed motion and the target future in one canonical coordinate frame, so the same transformer tokens can represent pose dynamics and global root movement together, and the global prediction is recovered by applying the inverse rotation and translation. Around this sit the GAT spatial embedding, which encodes skeleton adjacency, and the non-autoregressive transformer with causal masked temporal self-attention, relative position representations, and a shared-attention decoder, which together decode the whole output horizon in a single step.

What would settle it

Take a held-out set of sequences in which the person stands still for the full observation window and only starts moving after prediction time; if the two-point heading from Eq. (2) is used, the predicted global trajectory should show a large rotation error caused by the unstable angle, and this error should shrink when the heading is replaced by a longer-window or future-consistent estimate.

Watch

Extended reading notes

Core claim

The central claim is that separating the root joint from the remaining joints, as prior decoupled pose-and-trajectory methods do, is unnecessary: a transformer can output every joint including the root in global coordinates if the training sequences are first brought into a shared canonical frame. The motion transformation does this by translating each sequence so the root at the last observed frame sits at the origin and rotating the whole sequence so the root's recent motion direction points along the positive x-axis; prediction happens in that frame and the result is rotated and translated back. With a graph attention network supplying spatial skeleton structure and a non-autoregressive decoder predicting the full horizon in one pass, the resulting UPTor model achieves 0.51/0.74 m pose ADE/FDE and 0.12/0.25 m trajectory ADE/FDE on Human3.6M across all 15 actions, and 0.39/0.55 m pose and 0.13/0.26 m trajectory errors on DARKO, where it beats STPOTR. It does this with 23.2 million transformer parameters versus STPOTR's 43.3 million and a 17 ms forward pass, and the ablation shows the transformation makes trajectory predictions invariant to random global translations and rotations.

Load-bearing premise

The method assumes the heading measured between the last two input frames is a reliable reference direction for the whole prediction, which can fail when the person is nearly standing still, turning, or moving off the ground plane.

Editorial extensions

If this is right

  • A single forward pass of about 17 ms yields both a full-body pose sequence and a global trajectory, making the model usable as a real-time perception module on a mobile robot.
  • Because pose and trajectory are learned jointly rather than decoupled, trajectory predictions can draw on pose dynamics as context, which the paper credits for its trajectory accuracy.
  • The motion transformation makes predictions invariant to where and in which direction a person is moving relative to the robot, as shown by the translation and rotation ablation on DARKO.
  • On navigation-heavy data (DARKO, locomotion subset of CMU-Mocap), the unified model reports lower pose and trajectory errors than the decoupled STPOTR baseline.
  • The DARKO dataset, captured from a robot's egocentric view with 17 subjects and 508 navigation trials, provides a new benchmark for human-aware navigation prediction.

Reading between the lines

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

  • The motion transformation is stated to be applicable to any local pose prediction method, so the same canonical-frame trick could be grafted onto root-relative pose forecasters to give them global trajectory output without redesigning their decoders.
  • The heading used for orientation normalization is a two-point estimate over an unspecified interval delta; if that interval is not ablated, the method's claimed rotation invariance is only as strong as this estimate for near-stationary or turning motion.
  • The egocentric DARKO dataset could support richer contextual prediction, such as conditioning on the robot's own motion or static scene layout, since the robot's onboard sensor provides exactly the perspective a navigating robot needs.
  • A natural stress test is multi-person prediction, since the paper's single-person formulation would need to resolve which person's root anchors the canonical frame when several people are in view.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This paper proposes UPTor, a unified architecture for 3D full-body pose dynamics and root-joint trajectory prediction in a global coordinate frame. The method first normalizes each motion sequence by translating the last observed root position to the origin and rotating the motion direction to the positive x-axis (the 'motion transformation'), then uses a graph attention network for spatial joint embedding, sinusoidal spatial-temporal positional encodings, and a compact non-autoregressive transformer with shared attention to decode the entire future pose-and-trajectory sequence in one pass. The output is transformed back to global coordinates using the same translation vector and rotation angle. The authors evaluate on Human3.6M, a locomotion-filtered subset of CMU-Mocap, and a newly collected egocentric navigation dataset DARKO, and report ADE/FDE for pose joints and root trajectory, runtime, and ablations. The main claims are that the unified formulation avoids the decoupling of root and pose joints, and that the model is compact, real-time, and accurate for navigational human motion.

Significance. If the results hold, UPTor is a useful step for human-aware robot navigation: it offers a single non-autoregressive model for joint pose and trajectory forecasting, a transformation that is straightforward and applicable to other local pose predictors, and a new egocentric navigation dataset. The derivation is not circular: the translation vector v and rotation angle theta in Eqs. (1)-(2) are computed from observed input poses, not from ground-truth future poses, and I found no step in which the predicted output is reduced to a fitted constant. The main weaknesses are in the evaluation protocol rather than in the method's internal consistency; the navigation claims currently rest on subject-overlapping splits and on point estimates with no uncertainty quantification. Strengths of the manuscript include the clear formulation of the transformation, the explicit architectural description, and the inclusion of an ablation study.

major comments (3)
  1. [IV-A (Datasets) and Table II] The navigation claims rest on evaluation splits that are not subject-disjoint. For DARKO, the test set draws all trials from one actor and at least one trial from the other 16 actors, while the training set contains the remaining trials of those same 16 actors; for CMU-Mocap, training is drawn from all sampled subjects and the test set contains unseen trials of activities from these subjects. A transformer can exploit subject-specific gait, body proportions, and habitual motion style, so the ADE/FDE advantages reported in Table II may reflect identity leakage rather than the unified formulation. The H3.6M experiment does use held-out subjects S9/S11, but as the authors themselves note in Sec. IV-B, that evaluation set is dominated by static activities. I ask for subject-exclusive splits for DARKO and CMU-Mocap, or per-subject results showing that the advantage holds for the held-out actor, in addition to a navigation-focused evaluation on held-out subjects.
  2. [III-B, Eq. (2)] The orientation-invariance step depends on a two-point heading estimate computed from root positions at T1 and T1-delta, but delta is never specified and never ablated. For near-stationary poses, for turns occurring between T1-delta and T1, or for the wavy and deviating walks emphasized in DARKO, this estimate is noisy or ill-conditioned; since every input frame is rotated by theta and the predicted global trajectory is rotated back by the same theta, the final global output inherits the heading error. The paper should state delta, report the sensitivity of ADE/FDE to delta, and consider a more robust heading estimate such as an average over a longer observed window or a fit to the observed trajectory.
  3. [Table I and Sec. IV-B] The quantitative evidence for 'real-time and accurate' is weaker than the prose suggests. In Table I, the DLow/DMMGAN/STPOTR baseline errors are copied from [3] rather than re-measured; the inference-time column I is taken from the STPOTR paper's hardware, and the R values for baselines are scaled rather than measured. All metrics are single point estimates with no confidence intervals or repeated-seed evaluation, and several differences are small (e.g., UPTor 0.51/0.74 versus STPOTR 0.50/0.75 for ADE/FDE_Po on H3.6M). Without uncertainty quantification or re-running the baselines under the same protocol, the comparisons in Table I do not fully support the claimed accuracy and runtime advantage over STPOTR.
minor comments (4)
  1. [III-B (Loss)] The loss normalization appears incorrect: the sum runs from T1+1 to T1+T2, which contains T2 terms, but the prefactor is written as 1/(T2-T1-1). This should presumably be 1/T2, or 1/(T2-1) if the first predicted frame is handled differently; please correct and verify.
  2. [IV-A (Datasets)] The text contains small typos: 'at least one trail' should be 'at least one trial,' and 'such as including slow walking' should be 'such as slow walking.' Please also define T1 and T2 once in Sec. III-A and use them consistently in the loss expression.
  3. [Fig. 2] Figure 2 is very dense, and the decoder query initialization (repeating the encoder output for the last input pose) is not clearly visible; please enlarge the figure, add a legend for the shaded blocks, and explicitly annotate the back-transformation branch.
  4. [III-B (Motion Transformation)] The back-transformation is described only verbally; please give the explicit inverse equations (translate by -v and rotate by +theta) so that readers can verify that the global-coordinate output is correctly recovered.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the motion transformation is an input-derived normalization, and the self-citations are background material rather than load-bearing evidence.

full rationale

The derivation chain is self-contained. In Section III-B, the motion transformation normalizes each sequence using v = -j_root(T1) and the angle theta from Eq. (2); both quantities are computed from observed input poses only, not from ground-truth future poses and not from any fitted parameter. The inverse transform uses the same observed v and theta, so the predicted global trajectory is the model's output mapped through an invertible, input-dependent normalization rather than a quantity forced to equal its own input. Table III confirms the intended invariance property: with the transformation, errors are identical across test-time translations and rotations by construction, while without it they are not; this is a mathematical consequence of the normalization, not a circular reduction. The self-citations in the paper are [1] (a trajectory-prediction survey), [22] (a downstream MPC application), and [35] (the off-the-shelf MeTRAbs pose estimator used as perception input). None of these supplies the load-bearing claim that UPTor unifies pose and trajectory prediction; the architecture is built on external, independently published components such as Transformer [25], POTR [16], and GAT [39]. The limitations section honestly states that the method does not reach state-of-the-art performance on static activities and notes deployment issues with missing frames; these are performance caveats, not circular steps. The shared-subject DARKO and CMU-Mocap test splits are an evaluation-validity concern, but they do not make any prediction reduce to its inputs. No equation in the paper is equivalent to its own input by construction, so no circularity is identified.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no physical or ontological entities; DARKO is a benchmark dataset, not a postulated entity. The scientific load is carried by standard network components plus hand-set hyperparameters and domain assumptions about planar human navigation and pose-estimation quality. The least scrutinized hand-set quantity is delta, which is never assigned a value or ablated.

free parameters (3)
  • delta (motion direction lookback interval) = not reported
    Used in Eq. 2 to compute theta from root positions at T1 and T1-delta; hand-chosen and never ablated or given a numerical value.
  • Jdim (GAT joint embedding dimension) = 32
    Set to 32 for all datasets in Section IV-C; it directly controls the transformer dimension D = N * Jdim.
  • Training hyperparameters (learning rate, weight decay, epochs) = 1e-5, 1e-5, 20/125/50
    AdamW settings in Section IV-C are chosen per dataset without sensitivity analysis, and the number of transformer layers and attention heads is not reported.
assumptions (4)
  • domain assumption Human navigation trajectories are approximately planar, so rotation around the z-axis is sufficient to align motion direction.
    Eq. 2 and the rotation matrix apply only yaw; pitch and roll of the root trajectory are ignored.
  • domain assumption Instantaneous heading from root positions at T1-delta and T1 is a stable reference for the future horizon.
    Eq. 2 anchors the whole predicted sequence on this two-point direction; static or turning motions make theta unstable.
  • domain assumption DARKO ground-truth poses from the MeTRAbs estimator are accurate enough to train and evaluate a motion predictor.
    Section IV-A uses MeTRAbs absolute 3D pose estimates as labels without independent validation on this robot view.
  • standard math Multi-head attention and graph attention networks function as described in the cited literature.
    The model relies on standard transformer and GAT machinery without new theory.

how reviews work

0 comments
Cite this review

Pith. "Pith review of UPTor: Unified 3D Human Pose Dynamics and Trajectory Prediction for Human-Robot Interaction." pith.science (2026). https://pith.science/paper/LU6TTS3S

@misc{pith2026250514866,
  author       = {Pith},
  title        = {Pith review of: UPTor: Unified 3D Human Pose Dynamics and Trajectory Prediction for Human-Robot Interaction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LU6TTS3S}},
  note         = {Machine review of arXiv:2505.14866}
}
read the original abstract

We introduce a unified approach to forecast the dynamics of human keypoints along with the motion trajectory based on a short sequence of input poses. While many studies address either full-body pose prediction or motion trajectory prediction, only a few attempt to merge them. We propose a motion transformation technique to simultaneously predict full-body pose and trajectory key-points in a global coordinate frame. We utilize an off-the-shelf 3D human pose estimation module, a graph attention network to encode the skeleton structure, and a compact, non-autoregressive transformer suitable for real-time motion prediction for human-robot interaction and human-aware navigation. We introduce a human navigation dataset ``DARKO'' with specific focus on navigational activities that are relevant for human-aware mobile robot navigation. We perform extensive evaluation on Human3.6M, CMU-Mocap, and our DARKO dataset. In comparison to prior work, we show that our approach is compact, real-time, and accurate in predicting human navigation motion across all datasets. Result animations, our dataset, and code will be available at https://nisarganc.github.io/UPTor-page/

Figures

Figures reproduced from arXiv: 2505.14866 by the authors.

Figure 1
Figure 1. Human key points estimated from the robot’s perception stack and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. UPTor: Unified 3D Human Pose Dynamics and Trajectory Prediction Transformer [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Left: A top-down view of four color-coded motion sequences, with their corresponding transformed sequence encircled. Faded colors mark the motion start. Right: Transformation of a single sequence. Original motion is represented by a green trajectory and skeleton. Transformation parameters include the angle θ between the motion direction and the positive x-axis, and the translation vector v at T1. translated poses P … view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Distribution of locomotion velocities and participants’ heights in [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 6
Figure 6. Figure 6: DARKO dataset predictions across 2-second horizon at 16Hz [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: CMU-Mocap dataset predictions across 1-second horizon at 10Hz [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 40 canonical work pages

  1. [3]

    STPOTR: Simultaneous Human Trajectory and Pose Prediction Using a Non-Autoregressive Transformer for Robot Follow-Ahead

    M. Mahdavian, P. Nikdel, M. TaherAhmadi, and M. Chen. “STPOTR: Simultaneous Human Trajectory and Pose Prediction Using a Non-Autoregressive Transformer for Robot Follow-Ahead”. In: Proc. of the IEEE Int. Conf. on Robotics and Automation (ICRA) . IEEE. 2023, pp. 9959–9965

    work page 2023

  2. [1]

    Human motion trajectory prediction: A survey

    A. Rudenko, L. Palmieri, M. Herman, K. M. Kitani, D. M. Gavrila, and K. O. Arras. “Human motion trajectory prediction: A survey”. In: Int. J. of Robotics Research 39.8 (2020), pp. 895–935

    work page 2020

  3. [2]

    Proactive Model Predictive Control with Multi- Modal Human Motion Prediction in Cluttered Dynamic Environ- ments

    L. Heuer, L. Palmieri, A. Rudenko, A. Mannucci, M. Magnusson, and K. O. Arras. “Proactive Model Predictive Control with Multi- Modal Human Motion Prediction in Cluttered Dynamic Environ- ments”. In: Proc. of the IEEE Int. Conf. on Intell. Robots and Syst. (IROS). 2023

    work page 2023

  4. [4]

    Anticipating many futures: Online human motion prediction and generation for human- robot interaction

    J. Bütepage, H. Kjellström, and D. Kragic. “Anticipating many futures: Online human motion prediction and generation for human- robot interaction”. In: Proc. of the IEEE Int. Conf. on Robotics and Automation (ICRA) . IEEE. 2018, pp. 4563–4570

    work page 2018

  5. [5]

    Disentangling human dynamics for pedestrian locomotion forecast- ing with noisy supervision

    K. Mangalam, E. Adeli, K.-H. Lee, A. Gaidon, and J. C. Niebles. “Disentangling human dynamics for pedestrian locomotion forecast- ing with noisy supervision”. In: Proc. of the IEEE Conf. on Comp. Vis. and Pat. Rec. (CVPR) . 2020, pp. 2784–2793

    work page 2020

  6. [6]

    Future Frame Prediction Network for Human Fall Detection in Surveillance Videos

    S. Li and X. Song. “Future Frame Prediction Network for Human Fall Detection in Surveillance Videos”. In: IEEE Sensors Journal (2023)

    work page 2023

  7. [7]

    Deep neural networks enable quantitative movement analysis using single-camera videos

    Ł. Kidzi ´nski, B. Yang, J. L. Hicks, A. Rajagopal, S. L. Delp, and M. H. Schwartz. “Deep neural networks enable quantitative movement analysis using single-camera videos”. In: Nature com- munications 11.1 (2020), p. 4054

    work page 2020

  8. [8]

    Social LSTM: Human trajectory prediction in crowded spaces

    A. Alahi, K. Goel, V . Ramanathan, A. Robicquet, L. Fei-Fei, and S. Savarese. “Social LSTM: Human trajectory prediction in crowded spaces”. In: Proc. of the IEEE Conf. on Comp. Vis. and Pat. Rec. (CVPR). 2016, pp. 961–971

    work page 2016

Show all 43 references
  1. [9]

    Social GAN: Socially Acceptable Trajectories With Generative Adversarial Networks

    A. Gupta, J. Johnson, L. Fei-Fei, S. Savarese, and A. Alahi. “Social GAN: Socially Acceptable Trajectories With Generative Adversarial Networks”. In: Proc. of the IEEE Conf. on Comp. Vis. and Pat. Rec. (CVPR). 2018

    2018

  2. [10]

    Social ways: Learning multi-modal distributions of pedestrian trajectories with GANs

    J. Amirian, J.-B. Hayet, and J. Pettré. “Social ways: Learning multi-modal distributions of pedestrian trajectories with GANs”. In: Proc. of the IEEE Conf. on Comp. Vis. and Pat. Rec. (CVPR) Workshops. 2019

    2019

  3. [11]

    Multi-Agent Tensor Fusion for Contextual Trajectory Prediction

    T. Zhao, Y . Xu, M. Monfort, W. Choi, C. Baker, Y . Zhao, Y . Wang, and Y . N. Wu. “Multi-Agent Tensor Fusion for Contextual Trajectory Prediction”. In: Proc. of the IEEE Conf. on Comp. Vis. and Pat. Rec. (CVPR). 2019, pp. 12126–12134

    2019

  4. [12]

    Trajec- tron++: Dynamically-feasible trajectory forecasting with heteroge- neous data

    T. Salzmann, B. Ivanovic, P. Chakravarty, and M. Pavone. “Trajec- tron++: Dynamically-feasible trajectory forecasting with heteroge- neous data”. In: European Conference on Computer Vision. Springer. 2020, pp. 683–700

    2020

  5. [13]

    Wang and A

    Z. Wang and A. H. Qureshi. AnyPose: Anytime 3D Human Pose F orecasting via Neural Ordinary Differential Equations. 2023. arXiv: 2309.04840 [cs.CV]

    2023 arXiv

  6. [14]

    T. Ma, Y . Nie, C. Long, Q. Zhang, and G. Li. Progressively Generating Better Initial Guesses Towards Next Stages for High- Quality Human Motion Prediction . 2022. arXiv: 2203.16051

    2022 arXiv

  7. [15]

    W. Guo, Y . Du, X. Shen, V . Lepetit, X. Alameda-Pineda, and F. Moreno-Noguer. Back to MLP: A Simple Baseline for Human Motion Prediction. 2022. arXiv: 2207.01567 [cs.CV]

    2022 arXiv

  8. [16]

    Pose transformers (POTR): Human motion prediction with non- autoregressive transformers

    A. Martínez-González, M. Villamizar, and J.-M. Odobez. “Pose transformers (POTR): Human motion prediction with non- autoregressive transformers”. In: Proc. of the IEEE Conf. on Comp. Vis. and Pat. Rec. (CVPR) . 2021, pp. 2276–2284

    2021

  9. [17]

    Multi-Person 3D Motion Prediction with Multi-Range Transformers

    J. Wang, H. Xu, M. Narasimhan, and X. Wang. “Multi-Person 3D Motion Prediction with Multi-Range Transformers”. In: Advances in Neural Information Processing Systems . Ed. by A. Beygelzimer, Y . Dauphin, P. Liang, and J. W. Vaughan. 2021

    2021

  10. [18]

    Transformer net- works for trajectory forecasting

    F. Giuliari, I. Hasan, M. Cristani, and F. Galasso. “Transformer net- works for trajectory forecasting”. In: Proc. of the IEEE Int. Conf. on Pattern Recognition. IEEE. 2021, pp. 10335–10342

    2021

  11. [19]

    Long-term human motion prediction with scene context

    Z. Cao, H. Gao, K. Mangalam, Q.-Z. Cai, M. V o, and J. Ma- lik. “Long-term human motion prediction with scene context”. In: Proc. of the Europ. Conf. on Comp. Vision (ECCV) . Springer. 2020, pp. 387–404

    2020

  12. [20]

    DMMGAN: Diverse Multi Motion Prediction of 3D Human Joints using Attention-Based Gener- ative Adversarial Network

    P. Nikdel, M. Mahdavian, and M. Chen. “DMMGAN: Diverse Multi Motion Prediction of 3D Human Joints using Attention-Based Gener- ative Adversarial Network”. In: 2023 IEEE International Conference on Robotics and Automation (ICRA) . IEEE. 2023, pp. 9938–9944

    2023

  13. [21]

    Human-robot co-navigation using anticipatory indicators of human walking motion

    V . V . Unhelkar, C. Pérez-D’Arpino, L. Stirling, and J. A. Shah. “Human-robot co-navigation using anticipatory indicators of human walking motion”. In: Proc. of the IEEE Int. Conf. on Robotics and Automation (ICRA) . 2015, pp. 6183–6190

    2015

  14. [22]

    Efficient Context-Aware Model Predictive Control for Human-Aware Navigation

    E. Stefanini, L. Palmieri, A. Rudenko, T. Hielscher, T. Linder, and L. Pallottino. “Efficient Context-Aware Model Predictive Control for Human-Aware Navigation”. In: IEEE Robotics and Automation Letters. 2024

    2024

  15. [23]

    Dlow: Diversifying latent flows for diverse human motion prediction

    Y . Yuan and K. Kitani. “Dlow: Diversifying latent flows for diverse human motion prediction”. In: Proc. of the Europ. Conf. on Comp. Vision (ECCV). Springer. 2020, pp. 346–364

    2020

  16. [24]

    Learning decoupled representations for human pose forecasting

    B. Parsaeifard, S. Saadatnejad, Y . Liu, T. Mordan, and A. Alahi. “Learning decoupled representations for human pose forecasting”. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021, pp. 2294–2303

    2021

  17. [25]

    Attention is all you need

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin. “Attention is all you need”. In: Advances in Neural Inf. Proc. Syst. (NIPS) 30 (2017)

    2017

  18. [26]

    Human3.6M: Large Scale Datasets and Predictive Methods for 3D Human Sensing in Natural Environments

    C. Ionescu, D. Papava, V . Olaru, and C. Sminchisescu. “Human3.6M: Large Scale Datasets and Predictive Methods for 3D Human Sensing in Natural Environments”. In: IEEE Trans. on Patt. Anal. and Mach. Intell. (PAMI) 36.7 (2014), pp. 1325–1339

    2014

  19. [27]

    Accessed: 01-09-

    CMU Graphics Lab Motion Capture Database . Accessed: 01-09-

  20. [28]

    Barquero, S

    G. Barquero, S. Escalera, and C. Palmero. BeLFusion: Latent Dif- fusion for Behavior-Driven Human Motion Prediction . 2023. arXiv: 2211.14304 [cs.CV] . URL: https://arxiv.org/abs/ 2211.14304

    2023 arXiv

  21. [29]

    S. Tian, M. Zheng, and X. Liang. TransFusion: A Practical and Effective Transformer-based Diffusion Model for 3D Human Motion Prediction. 2023. arXiv: 2307.16106 [cs.RO]

    2023 arXiv

  22. [30]

    Simple baseline for single human motion forecasting

    C. Wang, Y . Wang, Z. Huang, and Z. Chen. “Simple baseline for single human motion forecasting”. In: Proc. of the IEEE Int. Conf. on Computer Vision (ICCV) . 2021, pp. 2260–2265

    2021

  23. [31]

    Socially and contextually aware human motion and pose fore- casting

    V . Adeli, E. Adeli, I. Reid, J. C. Niebles, and H. Rezatofighi. “Socially and contextually aware human motion and pose fore- casting”. In: IEEE Robotics and Automation Letters 5.4 (2020), pp. 6033–6040

    2020

  24. [32]

    Tripod: Human trajectory and pose dynamics forecasting in the wild

    V . Adeli, M. Ehsanpour, I. Reid, J. C. Niebles, S. Savarese, E. Adeli, and H. Rezatofighi. “Tripod: Human trajectory and pose dynamics forecasting in the wild”. In: Proc. of the IEEE Int. Conf. on Computer Vision (ICCV). 2021, pp. 13390–13400

    2021

  25. [33]

    Recurrent network models for human dynamics

    K. Fragkiadaki, S. Levine, P. Felsen, and J. Malik. “Recurrent network models for human dynamics”. In: Proc. of the IEEE Int. Conf. on Computer Vision (ICCV) . 2015, pp. 4346–4354

    2015

  26. [34]

    On human motion prediction using recurrent neural networks

    J. Martinez, M. J. Black, and J. Romero. “On human motion prediction using recurrent neural networks”. In: Proc. of the IEEE Conf. on Comp. Vis. and Pat. Rec. (CVPR) . 2017, pp. 2891–2900

    2017

  27. [35]

    MeTRAbs: metric-scale truncation-robust heatmaps for absolute 3d human pose estimation

    I. Sárándi, T. Linder, K. O. Arras, and B. Leibe. “MeTRAbs: metric-scale truncation-robust heatmaps for absolute 3d human pose estimation”. In: IEEE Transactions on Biometrics, Behavior , and Identity Science 3.1 (2020), pp. 16–30

    2020

  28. [36]

    C. Xu, R. T. Tan, Y . Tan, S. Chen, Y . G. Wang, X. Wang, and Y . Wang.EqMotion: Equivariant Multi-agent Motion Prediction with Invariant Interaction Reasoning . 2023. arXiv: 2303.10876

    2023 arXiv

  29. [37]

    Tevet, S

    G. Tevet, S. Raab, B. Gordon, Y . Shafir, D. Cohen-Or, and A. H. Bermano. Human Motion Diffusion Model . 2022. arXiv: 2209 . 14916 [cs.CV]

    2022

  30. [38]

    S. Xu, Z. Li, Y .-X. Wang, and L.-Y . Gui. InterDiff: Generating 3D Human-Object Interactions with Physics-Informed Diffusion . 2023. arXiv: 2308.16905 [cs.CV]

    2023 arXiv

  31. [39]

    Graph Attention Networks

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y . Bengio. “Graph Attention Networks”. In: International Conference on Learning Representations, ICLR . 2018

    2018

  32. [40]

    Self-Attention with Rela- tive Position Representations

    P. Shaw, J. Uszkoreit, and A. Vaswani. “Self-Attention with Rela- tive Position Representations”. In: North American Chapter of the Association for Computational Linguistics . 2018

    2018

  33. [41]

    Decoupled Weight Decay Regular- ization

    I. Loshchilov and F. Hutter. “Decoupled Weight Decay Regular- ization”. In: International Conference on Learning Representations, ICLR. 2019

    2019

  34. [42]

    Azure Kinect DK

    Microsoft. Azure Kinect DK . https : / / www . microsoft . com / en - us / d / azure - kinect - dk / 8pp5vxmd9nhq ? activetab = pivot : overviewtab. Accessed: 14-Sep-2024. 2024

    2024

  35. [2023]

    URL: http://mocap.cs.cmu.edu/

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.