PFM-HR: Pose Flow Matching for Humanoid Robots

arXiv:2608.03227 · cs.RO · Submitted 2026-08-04 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "PFM-HR: Pose Flow Matching for Humanoid Robots".

Dev: Here's the gist of the paper. Motion priors improve reinforcement learning for physics-based humanoid tracking by guiding policy optimization toward behaviors supported by demonstrations The paper presents Pose Flow Matching for Humanoid Robots (PFM-HR), a reusable flow matching prior trained directly on large-scale unordered pose data This prior introduces the Pose Geometry Score (PGS), which quantifies how joint coordinate changes during rollouts align with the local geometry of pose variation captured by the prior Using PGS to modulate the tracking reward guides policy exploration toward structured pose changes while keeping the prior frozen across tracking tasks.

Rosa: Experiments demonstrate that PFM-HR improves both single-motion and general motion tracking, especially for highly dynamic motions.

Dev: Problem Context and Motivation Humanoid robots hold considerable promise for operating in human-centered environments where their humanlike morphology allows them to use the same spaces, tools, and interaction patterns as people Dynamic feasibility alone does not specify how the joints should vary together to produce coordinated humanlike motion.

Rosa: The authors address a gap between efficient pose-wise priors, which do not evaluate how joints change together, and temporal priors, which require ordered data and heavier online inference.

Title and authors: Rosa: So this paper, PFM-HR: Pose Flow Matching for Humanoid Robots, it’s talking about how we can get better control for these robots by using pose information in a new way. It seems like they are trying to solve a problem where just having a good pose prior isn't enough because it doesn't tell you how the joints should move together.

Dev: Right, that’s the core issue they are tackling. We have these great motion priors, but they don't actually evaluate how joint coordinates change in coordination during a movement. This paper introduces something called Pose Flow Matching for Humanoid Robots as a reusable prior trained directly on large-scale unordered pose data to guide policy optimization towards behaviors that are structured in terms of joint changes.

Taro: So, it’s about moving away from just having static pose knowledge and into understanding the dynamics of how the joints actually move together across time, which is pretty important when you're simulating something dynamic like a backflip or a double kong.

Rosa: Exactly. They introduce this thing called the Pose Geometry Score, or PGS, which they use to measure if the joint coordinate changes during a simulation rollout line up with the local geometry of pose variation that their prior has learned from all that unordered data.

Dev: That PGS is what they use to modulate the tracking reward, and it’s designed so that it guides policy exploration toward structured pose changes while keeping this prior frozen across different tracking tasks. That freezing part is key because you don't want to retrain a big model every time you switch from one task to another.

Taro: I wonder if that freezing part is actually the trick here, or if it means we miss out on learning new coordination patterns as the robot interacts with different environments.

Rosa: That’s a fair question, Taro. The paper suggests that by using this PGS score to adjust the reward, you reduce learning signals from those change patterns that aren't supported well by the prior, which is useful for keeping things stable and focused on what actually matters for tracking.

Dev: And in terms of performance, they show that PFM-HR improves both single-motion tracking and general motion tracking, especially when the motions are highly dynamic. They also claim it converges faster on specific tasks like backflip and double kong compared to other methods.

Taro: Faster convergence sounds good for real-world deployment because it means less time spent in simulation figuring out how to move correctly before you even get to testing on hardware.

Title and authors: Rosa: It’s not just about speed, though they also show that in general motion tracking, PFM-HR achieves the lowest position and rotation errors at all three horizons when using the same backbone as other methods. That suggests it finds a better balance between learning speed and actual accuracy for complex movements.

Dev: The way they build this training signal involves a linear Flow Matching path that interpolates from a Gaussian source distribution at time zero to the pose distribution at time one, and the velocity of that flow is defined by the difference between where you are in pose space and some epsilon value.

Taro: So it’s not just feeding it random poses; they are using this specific matching process to teach the model about how these joint coordinate changes relate to each other geometrically across a sequence of poses.

Rosa: That’s right, and the way they characterize the rollout update ending at step k is by looking at that finite-difference joint coordinate velocity, which records those changes induced by the policy over one control interval.

Dev: The Pose Geometry Score is defined as something like the square of the norm of the Jacobian of that flow with respect to those joint coordinates, multiplied by a term related to the environment’s local pose variation.

Taro: That sounds mathematically intensive, but if it really tells you whether the joint changes align with what they learned from all those unordered poses, then it should give the policy something meaningful to optimize for.

Rosa: It does, and this is where I see its main practical benefit: it guides policy exploration toward structured pose changes while keeping the prior frozen across tracking tasks. This means one model can be useful for many different tracking challenges without needing a massive overhaul every time you change the goal.

Dev: They also mentioned a reward modulation mechanism where the final reward combines a task reward with this geometry-aware term, and they attenuate the tracking reward below a certain threshold, so it helps focus on the right kind of changes.

Taro: But I do have to point out their limitation here—they say this prior captures local co-variation in the marginal pose distribution rather than temporal dynamics, which means it doesn't really capture the likelihood or direction of temporal transitions between poses.

Rosa: That’s a crucial caveat, Taro. Because they trained on unordered data and the score is invariant to direction reversal at a fixed query pose, it can't actually assess the ordering of those movements in time.

Dev: And because it relies on that local geometry learned from unordered poses, its guidance might be less informative for poses that are far away from that original training distribution. So, it’s not a universal solution for every possible motion scenario.

Title and authors: Taro: I guess the real test will be how robust it is when things go wrong in a real-world setting, like when the world misbehaves or we hit an unexpected physical constraint.

Rosa: That’s what we want to see outside the lab, Taro. They also show that the normalized PGS measures how much of the original score is retained after some perturbation, and this score remains within one point six percent of its unperturbed value even when things are messed with a bit.

Dev: And they suggest that perturbations that actually disrupt the inter-joint coordination pattern are expected to reduce this normalized score, decrease the calibrated percentile and lower the reward multiplier for those changes. So it’s robust to noise but sensitive to actual coordinated movement structure.

Taro: So, if we have a noisy sensor or a slightly unexpected physical interaction that messes up how the joints should move together, this score drops, which tells the policy it’s doing something structurally wrong in terms of its motion plan.

Rosa: That’s exactly what they are trying to achieve: giving the robot guidance based on structure rather than just brute-force tracking error. This paper, PFM-HR: Pose Flow Matching for Humanoid Robots, shows that local geometry learned from unordered poses can provide useful control guidance even without explicit temporal supervision.

Dev: And for the folks in real-time loops, I see the efficiency gain too; they found that PFM-HR requires fewer simulation samples to reach an error level of eighty percent across all skills compared to baselines, which is a big win for sample efficiency.

Taro: If it cuts down the number of simulation samples needed significantly, that translates directly into less time waiting for training and more time actually testing the control policies on the physical robot.

Rosa: So to wrap up, PFM-HR: Pose Flow Matching for Humanoid Robots gives us a reusable flow matching prior trained on unordered pose data with a Pose Geometry Score that modulates rewards to guide policy optimization toward structured pose changes while keeping the prior frozen across tasks. It improves sample efficiency and tracking accuracy especially for dynamic motions.

Dev: It’s about providing geometry-aware guidance that helps policies learn better coordination patterns, even though the prior is limited in capturing temporal ordering or likelihood of transitions between poses.

Taro: For me, it means we can deploy these policies on a humanoid robot with more confidence because they've been guided by a learned understanding of how the joints are supposed to move together geometrically.

Rosa: And that’s where we are heading next, looking at how this kind of geometric guidance translates into reliable control when the robot is actually out in the open and things get messy.

The paper's summary: Rosa: So, what we're looking at here is Pose Flow Matching for Humanoid Robots, or PFMHR. Basically, this paper introduces a new training method for these robots that uses pose data in a way that’s really guiding the policy optimization toward movements that look structurally sound.

Dev: Yeah, it takes large-scale unordered pose data and trains a flow matching prior directly on it. The main idea is to introduce something they call the Pose Geometry Score, or PGS.

Rosa: The PGS measures how well the joint coordinate changes during a simulation rollout match up with the geometric patterns that the prior learned from all those poses. It’s basically checking if what the robot is trying to do with its joints actually makes sense according to what it saw before.

Dev: And they use this score to modulate the tracking reward. They use it to guide policy exploration toward structured pose changes while keeping this prior frozen across different tasks, which means you don't have to retrain everything for every new tracking job.

Rosa: That’s a big deal because it separates the learning of how joints should move geometrically from the actual task of tracking a specific motion. It’s like giving the robot a built-in sense of what a good, coordinated movement looks like.

Dev: The experiments show that this PFMHR approach improves both single-motion tracking and general motion tracking, especially when the movements are very dynamic, like backflips or double kong. They also claim it converges faster on those specific skills.

Rosa: Faster convergence is important for real-time applications because it means less time spent in simulation before you get the robot running on hardware. And they say it finds a really good balance between how fast the policy learns and how accurate the tracking actually is, especially for those acrobatic motions.

Dev: They also have this reward modulation mechanism where they change the tracking reward based on that geometry score, making sure it only gets strong signals when the pose changes are structured in a way that matches their prior knowledge.

Rosa: Now, I do have to point out what they didn't cover much—this prior captures local co-variation in the marginal pose distribution rather than temporal dynamics. So, it doesn't really know the likelihood or the direction of transitions between poses in time.

Dev: That’s a limitation. Because they trained on unordered data and that score is invariant to direction reversal at a fixed query pose, it can't actually assess the ordering of those movements in time. It’s good for structure, but not necessarily for sequential motion planning.

Rosa: Exactly. So while this gets us better at getting the *shape* of the joint movement right, we still have to figure out the *sequence* and *timing* of those movements on our robots.

Dev: And they also mentioned that for prediction parameterization, direct clean pose prediction actually performs better on dynamic motions, particularly backflip, because it gives a better Jacobian response for a finite capacity model.

Rosa: So to wrap up this part—PFMHR gives us geometry-aware guidance that helps policies learn better coordination patterns without needing explicit temporal supervision. It’s about improving sample efficiency and tracking accuracy for dynamic stuff, but we still have to worry about those sequential timing issues.

Dev: And that leads us right into the next part of the discussion where we look at how this geometry-aware guidance actually plays out when things get messy in the real world and what that means for safety.

The paper's improvements: Rosa: So, we're looking at how PFMHR makes things better—it’s not just about getting a new prior, but about how they use that prior to improve the actual control system. They suggest using this geometry score to actively guide policy optimization toward those structured pose changes we talked about earlier.

Dev: Exactly. The paper shows that this guidance helps the policy explore the right kind of movements more efficiently, which is especially useful when we’re dealing with complex, dynamic stuff like backflips or double kong. It really improves sample efficiency on those kinds of motions and tracking accuracy for general motion tracking too.

Rosa: I mean, what this means for me out there on the field is that if my robot tries to do a weird, uncoordinated move, the reward signal from PFMHR will pull it back toward something that looks geometrically sensible according to what it learned from all that pose data.

Dev: And they found that this method actually improves long-horizon general motion tracking reliability by reducing position error by about seven point six percent over the other methods when you look at longer episodes. It seems like the geometry-aware guidance helps keep things stable over time, rather than just getting a quick fix for one frame.

Rosa: That's interesting, because it shows that this isn't just a quick boost to tracking; it’s making the overall learned behavior more robust for longer tasks. What about those limitations they mentioned?

Dev: They did flag something important: this prior doesn't capture temporal dynamics well. It doesn't really know the likelihood or the direction of transitions between poses, which means it’s not great for learning long sequences of movements where timing is everything.

Rosa: So, it’s like having a really good eye for structure in a single pose, but still needing something else to tell the robot *when* and *how* to transition smoothly from pose A to pose B. That’s what we need to figure out next.

Dev: Right. And they also showed that this score is quite robust; it stays within about one point six percent of its original value even when you add some noise or perturbations, as long as those perturbations don't totally break the coordination pattern between joints.

Rosa: So, the implication is that we can deploy these policies on real robots with more confidence because they’ve been steered by this geometric understanding, even if it doesn't have perfect knowledge of the entire timeline.

Dev: That’s a solid point. And for the controls engineers out there, it means we can use this frozen prior as a foundation for many different tracking tasks without having to rebuild our whole motion model every time we switch goals.

Conclusion: Rosa: So, to wrap up, PFMHR is this reusable flow matching prior that uses a Pose Geometry Score to guide policy optimization toward structured pose changes while keeping the prior frozen across tasks. It's essentially a way to give humanoid robots geometric intuition about how their joints should move together without needing explicit temporal supervision.

Dev: Exactly. The paper shows that this approach significantly boosts sample efficiency and tracking accuracy, particularly when we’re dealing with dynamic motions like backflips, making it a strong tool for training complex behaviors faster.

Rosa: It seems like the main implication is that we can use a single model trained with this prior for many different tracking tasks because the geometry part of it stays fixed while the policy learns to adapt to the specific goal.

Dev: That’s right. We get a more robust starting point for our policies, which means less time spent on trial and error when deploying on actual robots. But we still have that limitation we talked about—it doesn't capture the full temporal sequence of those movements.

Taro: I still think the lack of temporal modeling is a hurdle for true autonomy in complex, long-horizon scenarios where the timing of joint movements is everything.

Rosa: True. So while PFMHR gives us great geometry-aware guidance, we still need to figure out how to integrate that with better sequence prediction methods down the road.

Dev: Definitely. That leads us right into looking at papers that focus on trajectory-driven policy generation and how those can be combined with geometric priors for smoother, more temporally coherent control.

HKUST(GZ) · Noitom Robotics · SIGS Tsinghua University · Google

cs.RO

Submitted: 2026-08-04

Updated: 2026-11-05

Comments: 7 pages

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 98/100

The gist: Motion priors improve reinforcement learning for physics-based humanoid tracking by guiding policy optimization toward behaviors supported by demonstrations The paper presents Pose Flow Matching for…

Key concepts

Pose Flow Matching Prior (PFM-HR)
This is a reusable model trained directly on large collections of unordered humanoid poses. It learns the underlying distribution of poses, allowing it to provide geometric guidance during reinforcement learning without needing explicit temporal ordering or complex online inference for every task.
Pose Geometry Score (PGS)
The PGS measures how well the joint coordinate changes during a robot's movement match the local geometric patterns learned by the PFM-HR prior. A high score means the joint movements follow patterns strongly represented in the training data, helping guide policy exploration towards structured poses.
Reward Modulation Mechanism
The final reward signal is adjusted based on whether a pose change aligns with the learned geometry via PGS. This modulation reduces learning signals from poorly supported changes, ensuring that the robot learns to perform coordinated, physically plausible movements rather than random joint variations.

Terminology

Summary

Motion priors improve reinforcement learning for physics-based humanoid tracking by guiding policy optimization toward behaviors supported by demonstrations The paper presents Pose Flow Matching for Humanoid Robots (PFM-HR), a reusable flow matching prior trained directly on large-scale unordered pose data This prior introduces the Pose Geometry Score (PGS), which quantifies how joint coordinate changes during rollouts align with the local geometry of pose variation captured by the prior Using PGS to modulate the tracking reward guides policy exploration toward structured pose changes while keeping the prior frozen across tracking tasks. p. 0

Where the paper says it

PFMHR introduces the Pose Geometry Score (PGS), which quantifies how joint coordinate changes during rollouts align with the local geometry of pose variation captured by the prior. Using PGS to modulate the tracking reward guides policy exploration toward structured pose changes while keeping the prior frozen across tracking tasks.

Experiments demonstrate that PFM-HR improves both single-motion and general motion tracking, especially for highly dynamic motions. p. 0
Where the paper says it

Using PGS to modulate the tracking reward guides policy exploration toward structured pose changes while keeping the prior frozen across tracking tasks. Experiments demonstrate that PFM-HR improves both single-motion and general motion tracking, especially for highly dynamic motions.

Problem Context and Motivation

Humanoid robots hold considerable promise for operating in human-centered environments where their humanlike morphology allows them to use the same spaces, tools, and interaction patterns as people Dynamic feasibility alone does not specify how the joints should vary together to produce coordinated humanlike motion. The authors address a gap between efficient pose-wise priors, which do not evaluate how joints change together, and temporal priors, which require ordered data and heavier online inference. p. 0

Where the paper says it

This leaves a gap between efficient pose-wise priors, which do not evaluate how joints change together, and temporal priors, which require ordered data and heavier online inference. We address this gap with Pose Flow Matching for Humanoid Robots (PFM-HR), a reusable Flow Matching prior trained directly on individual humanoid poses, with the training corpus treated as an unordered collection [11].

How the Training Signal is Built

The PFM-HR prior is trained on individual humanoid poses sampled from a large-scale motion corpus treated as an unordered pose collection The training data are treated as an unordered pose collection, from which the model learns the marginal pose distribution pdata. The linear Flow Matching path interpolates between the Gaussian source distribution at t = 0 and the pose distribution at t = 1, with its conditional velocity being v = dzt/dt = x - epsilon. p. 2

Where the paper says it

(5) This path interpolates between the Gaussian source distribution at t = 0 and the pose distribution at t = 1. Its conditional velocity is v = dzt dt = x − ϵ.

The authors optimize the Flow Matching objective LFM(ϕ) = E[t,x,epsilon] h v - vϕ(zt, t) 2 2 i. p. 2
Where the paper says it

The predicted velocity is vϕ(zt, t) = xˆϕ(zt, t) − zt max(1 − t, δ) , (7) where δ > 0 stabilizes the denominator near t = 1. We optimize the Flow Matching objective LFM(ϕ) = Et,x,ϵ h ∥v − vϕ(zt, t)∥ 2 2 i .

Where the Gains Come From

The Pose Geometry Score (PGS) is defined as sPGS,k:= Jϕ,k dk 2 2 = Jϕ,kξenv,k 2 2 ξenv,k 2 2. A high PGS indicates that the relative joint changes align with co-variation modes strongly represented by the pose prior. p. 0

Where the paper says it

PFM-HR applies this Jacobian to the normalized jointcoordinate change between consecutive rollout poses and defines the squared Jacobian–vector response as the Pose Geometry Score (PGS). A high PGS indicates that the relative joint changes align with co-variation modes strongly represented by the pose prior.

This score is then used to modulate the tracking reward to reduce learning signals from weakly supported change patterns. p. 0
Where the paper says it

A high PGS indicates that the relative joint changes align with co-variation modes strongly represented by the pose prior. After reference-specific calibration, PGS modulates the tracking reward to reduce learning signals from weakly supported change patterns.

Experimental Results and Performance

Experiments on single-trajectory tracking show that ADD w/ PFM-HR performs best on the other six tasks, and PFM-HR converges faster on both Backflip and Double Kong. In general motion tracking, PFM-HR achieves the lowest position and rotation errors at all three horizons under the same ADD backbone. p. 4

Where the paper says it

Table II reports general motion tracking errors across different episode lengths. Under the same ADD backbone and training budget, PFM-HR achieves the lowest position and rotation errors at all three horizons.

The results show that PFM-HR provides the best balance between convergence and tracking accuracy, particularly on dynamic motions. p. 4
Where the paper says it

ADD w/ PFM-HR also achieves the lowest or tied-lowest positional error on eight tasks, although the margins are generally small. Overall, PFM-HR provides the best balance between convergence and tracking accuracy, particularly on dynamic motions.

Limitations of the Prior

PFM-HR captures local co-variation in the marginal pose distribution rather than temporal dynamics Because the training poses are unordered and PGS is invariant to direction reversal at a fixed query pose, it cannot assess the likelihood, direction, or ordering of temporal transitions. The reliability of PGS also depends on the coverage of the pose dataset, and its guidance may be less informative for poses far from the training distribution. p. 6

Where the paper says it

Consequently, opposite transitions may receive similar geometric evaluations even when only one is consistent with the intended motion. The reliability of PGS also depends on the coverage of the pose dataset, and its guidance may be less informative for poses far from the training distribution.

Reward Modulation Mechanism

The final reward is defined as rk = w G r G k + w T r T k rP k. The tracking reward is unchanged above pgood and maximally attenuated below pbad. p. 3

Where the paper says it

The final reward is rk = w Gr G k + w T r T k r P k . (18) The tracking reward is unchanged above pgood and maximally attenuated below pbad.

The authors compare PFM-HR to FMRecon, showing that PFM-HR provides larger improvements on dynamic motions. p. 5
Where the paper says it

Comparing ADD with FM-Recon evaluates the contribution of the frozen prior, while comparing FM-Recon with PFMHR assesses multi-level pose reconstruction against directional Jacobian scoring. Table III shows that PFM-HR provides larger improvements on dynamic motions.

Prediction Parameterization Comparison

Comparing vprediction and x-prediction shows that x-prediction performs better on dynamic motions, particularly Backflip Because the two variants share the same architecture and prior-query cost, the results suggest that direct clean-pose prediction provides a more effective Jacobian response for a finite-capacity model. The single JVP is used during policy training, drawing a fresh corruption sample at every rollout step, and therefore requires only one JVP per rollout step. p. 2

Where the paper says it

(12) Pose Geometry Score. At rollout step k, we construct a query using the same corruption parameterization as in pretraining: qek−1 = tevalqk−1 + (1 − teval)ϵk, ϵk ∼ N (0, I), (13) where teval ≤ 1 − δ is fixed during policy learning.

Robustness to Perturbations

The normalized PGS measures the fraction of the original score retained after perturbation In contrast, perturbations that progressively disrupt the inter-joint coordination pattern are expected to reduce RPGS, decrease the calibrated percentile and reward multiplier. PGS is nearly invariant to the overall magnitude of a pose change In contrast, perturbations that progressively disrupt the inter-joint coordination pattern are expected to reduce RPGS, decrease the calibrated percentile and reward multiplier. p. 2

Where the paper says it

The joint-coordinate change pattern is represented by dk = q ξenv,k ∥ξenv,k∥ 2 2 + εnum , (14) which retains the relative changes across joints while reducing sensitivity to their overall magnitude. We define the Pose Geometry Score (PGS) as sPGS,k := ∥Jϕ,kdk∥ 2 2 = ∥Jϕ,kξenv,k∥ 2 2 ∥ξenv,k∥ 2 2 + εnum .

The results suggest that PGS is robust to nuisance variations while remaining sensitive to the inter-joint coordination patterns.

Conclusion

PFM-HR presents a reusable Flow Matching pose prior for physics-based humanoid motion tracking that can be trained directly on large-scale unordered pose data Its Pose Geometry Score uses the denoiser Jacobian to evaluate whether policy-induced pose changes align with the joint co-variation patterns captured by the prior. By separating prior learning from policy optimization, PFM-HR allows a single pretrained model to provide geometry-aware guidance across different tracking tasks. p. 6

Where the paper says it

This score can be computed efficiently through a Jacobian–vector product and used to modulate tracking rewards while keeping the prior frozen. By separating prior learning from policy optimization, PFM-HR allows a single pretrained model to provide geometry-aware guidance across different tracking tasks.

Experiments demonstrate improved sample efficiency and tracking accuracy, particularly for dynamic and acrobatic motions, as well as compatibility with real-world humanoid deployment. p. 6
Where the paper says it

By separating prior learning from policy optimization, PFM-HR allows a single pretrained model to provide geometry-aware guidance across different tracking tasks. Experiments demonstrate improved sample efficiency and tracking accuracy, particularly for dynamic and acrobatic motions, as well as compatibility with real-world humanoid deployment.

The results show that the local geometry learned from unordered poses can provide useful control guidance even without explicit temporal supervision. p. 6
Where the paper says it

Experiments demonstrate improved sample efficiency and tracking accuracy, particularly for dynamic and acrobatic motions, as well as compatibility with real-world humanoid deployment. These results show that the local geometry learned from unordered poses can provide useful control guidance even without explicit temporal supervision.

References

[1] Arthur Allshire, Hongsuk Choi, Junyi Zhang, David McAllister, Anthony Zhang, Chung Min Kim, Trevor Darrell, Pieter Abbeel, Jitendra Malik, and Angjoo Kanazawa. Visual imitation enables contextual humanoid control. p. 7

Where the paper says it

REFERENCES [1] Arthur Allshire, Hongsuk Choi, Junyi Zhang, David McAllister, Anthony Zhang, Chung Min Kim, Trevor Darrell, Pieter Abbeel, Jitendra Malik, and Angjoo Kanazawa. Visual imitation enables contextual humanoid control.

In Proceedings of the Conference on Robot Learning (CoRL), 2025. p. 7
Where the paper says it

Visual imitation enables contextual humanoid control. In Proceedings of the Conference on Robot Learning (CoRL), 2025.

[2] Bones Studio. p. 7

Where the paper says it

In Proceedings of the Conference on Robot Learning (CoRL), 2025. [2] Bones Studio.

Ai datasets for machine learning and motion capture. p. 7
Where the paper says it

[2] Bones Studio. Ai datasets for machine learning and motion capture.

https://bones. p. 7
Where the paper says it

Ai datasets for machine learning and motion capture. https://bones.

studio/ai-datasets/, 2026. p. 7
Where the paper says it

https://bones. studio/ai-datasets/, 2026.

[3] Zixuan Chen, Mazeyu Ji, Xuxin Cheng, Xuanbin Peng, Xue Bin Peng, and Xiaolong Wang. p. 7

Where the paper says it

Accessed: 2026-05-06. [3] Zixuan Chen, Mazeyu Ji, Xuxin Cheng, Xuanbin Peng, Xue Bin Peng, and Xiaolong Wang.

Gmt: General motion tracking for humanoid whole-body control. p. 7
Where the paper says it

[3] Zixuan Chen, Mazeyu Ji, Xuxin Cheng, Xuanbin Peng, Xue Bin Peng, and Xiaolong Wang. Gmt: General motion tracking for humanoid whole-body control.

arXiv:2506. p. 7
Where the paper says it

Gmt: General motion tracking for humanoid whole-body control. arXiv:2506.

14770, 2025. p. 7
Where the paper says it

arXiv:2506. 14770, 2025.

[4] Alejandro Escontrela, Xue Bin Peng, Wenhao Yu, Tingnan Zhang, Atil Iscen, Ken Goldberg, and Pieter Abbeel. p. 7

Where the paper says it

14770, 2025. [4] Alejandro Escontrela, Xue Bin Peng, Wenhao Yu, Tingnan Zhang, Atil Iscen, Ken Goldberg, and Pieter Abbeel.

Adversarial motion priors make good substitutes for complex reward functions. p. 7
Where the paper says it

[4] Alejandro Escontrela, Xue Bin Peng, Wenhao Yu, Tingnan Zhang, Atil Iscen, Ken Goldberg, and Pieter Abbeel. Adversarial motion priors make good substitutes for complex reward functions.

In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 25–32. p. 7
Where the paper says it

Adversarial motion priors make good substitutes for complex reward functions. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 25–32.

IEEE, 2022. p. 7
Where the paper says it

Adversarial motion priors make good substitutes for complex reward functions. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 25–32.

[5] Yi Gu, Yukang Gao, Yangchen Zhou, Xingyu Chen, Yixiao Feng, Mingle Zhao, Yunyang Mo, Zhaorui Wang, Lixin Xu, and Renjing Xu. p. 7

Where the paper says it

IEEE, 2022. [5] Yi Gu, Yukang Gao, Yangchen Zhou, Xingyu Chen, Yixiao Feng, Mingle Zhao, Yunyang Mo, Zhaorui Wang, Lixin Xu, and Renjing Xu.

Hr: Pose distance fields for humanoid robots. p. 7
Where the paper says it

[5] Yi Gu, Yukang Gao, Yangchen Zhou, Xingyu Chen, Yixiao Feng, Mingle Zhao, Yunyang Mo, Zhaorui Wang, Lixin Xu, and Renjing Xu. Hr: Pose distance fields for humanoid robots.

arXiv preprint arXiv:2602. p. 7
Where the paper says it

Hr: Pose distance fields for humanoid robots. arXiv preprint arXiv:2602.

04851, 2026. p. 7
Where the paper says it

arXiv preprint arXiv:2602. 04851, 2026.

[6] Felix G Harvey, Mike Yurick, Derek Nowrouzezahrai, ´ and Christopher Pal. p. 7

Where the paper says it

04851, 2026. [6] Felix G Harvey, Mike Yurick, Derek Nowrouzezahrai, ´ and Christopher Pal.

Robust motion in-betweening. p. 7
Where the paper says it

[6] Felix G Harvey, Mike Yurick, Derek Nowrouzezahrai, ´ and Christopher Pal. Robust motion in-betweening.

ACM Transactions on Graphics (TOG), 39(4):60–1, 2020. p. 7
Where the paper says it

Robust motion in-betweening. ACM Transactions on Graphics (TOG), 39(4):60–1, 2020.

[7] Xialin He, Sirui Xu, Xinyao Li, Runpei Dong, Liuyu Bian, Yu-Xiong Wang, and Liang-Yan Gui. p. 7

Where the paper says it

ACM Transactions on Graphics (TOG), 39(4):60–1, 2020. [7] Xialin He, Sirui Xu, Xinyao Li, Runpei Dong, Liuyu Bian, Yu-Xiong Wang, and Liang-Yan Gui.

Ultra: Unified multimodal control for autonomous humanoid whole-body loco-manipulation. p. 7
Where the paper says it

[7] Xialin He, Sirui Xu, Xinyao Li, Runpei Dong, Liuyu Bian, Yu-Xiong Wang, and Liang-Yan Gui. Ultra: Unified multimodal control for autonomous humanoid whole-body loco-manipulation.

arXiv preprint arXiv:2603. p. 7
Where the paper says it

Ultra: Unified multimodal control for autonomous humanoid whole-body loco-manipulation. arXiv preprint arXiv:2603.

p. 7
Where the paper says it

Ultra: Unified multimodal control for autonomous humanoid whole-body loco-manipulation. arXiv preprint arXiv:2603.

03279, 2026. p. 7
Where the paper says it

arXiv preprint arXiv:2603. 03279, 2026.

[8] Yannan He, Garvita Tiwari, Tolga Birdal, Jan Eric Lenssen, and Gerard Pons-Moll. p. 7

Where the paper says it

03279, 2026. [8] Yannan He, Garvita Tiwari, Tolga Birdal, Jan Eric Lenssen, and Gerard Pons-Moll.

Nrdf: Neural riemannian distance fields for learning articulated pose priors. p. 7
Where the paper says it

[8] Yannan He, Garvita Tiwari, Tolga Birdal, Jan Eric Lenssen, and Gerard Pons-Moll. Nrdf: Neural riemannian distance fields for learning articulated pose priors.

In Conference on Computer Vision and Pattern Recognition (CVPR), 2024. p. 7
Where the paper says it

Nrdf: Neural riemannian distance fields for learning articulated pose priors. In Conference on Computer Vision and Pattern Recognition (CVPR), 2024.

[9] Tianhong Li and Kaiming He. p. 7

Where the paper says it

In Conference on Computer Vision and Pattern Recognition (CVPR), 2024. [9] Tianhong Li and Kaiming He.

Back to basics: Let denoising generative models denoise. p. 7
Where the paper says it

[9] Tianhong Li and Kaiming He. Back to basics: Let denoising generative models denoise.

arXiv preprint arXiv:2511. p. 7
Where the paper says it

Back to basics: Let denoising generative models denoise. arXiv preprint arXiv:2511.

p. 7
Where the paper says it

Back to basics: Let denoising generative models denoise. arXiv preprint arXiv:2511.

13720, 2025. p. 7
Where the paper says it

arXiv preprint arXiv:2511. 13720, 2025.

[10] Qiayuan Liao, Takara E. p. 7

Where the paper says it

13720, 2025. [10] Qiayuan Liao, Takara E.

Truong, Xiaoyu Huang, Yuman Gao, Guy Tevet, Koushil Sreenath, and C. p. 7
Where the paper says it

[10] Qiayuan Liao, Takara E. Truong, Xiaoyu Huang, Yuman Gao, Guy Tevet, Koushil Sreenath, and C.

Karen Liu. Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion, 2025. p. 7
Where the paper says it

Truong, Xiaoyu Huang, Yuman Gao, Guy Tevet, Koushil Sreenath, and C. Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion, 2025.

URL https://arxiv. p. 7
Where the paper says it

Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion, 2025. URL https://arxiv.

p. 7
Where the paper says it

Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion, 2025. URL https://arxiv.

p. 7
Where the paper says it

Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion, 2025. URL https://arxiv.

org/abs/2508. p. 7
Where the paper says it

URL https://arxiv. org/abs/2508.

08241.

[11] Yaron Lipman, Ricky T. p. 7

Where the paper says it

org/abs/2508. [11] Yaron Lipman, Ricky T.

Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. p. 7
Where the paper says it

[11] Yaron Lipman, Ricky T. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le.

Flow matching for generative modeling. p. 7
Where the paper says it

Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling.

In The Eleventh International Conference on Learning Representations, 2023. p. 7
Where the paper says it

Flow matching for generative modeling. In The Eleventh International Conference on Learning Representations, 2023.

URL https://openreview. p. 7
Where the paper says it

In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.

net/forum?id=PqvMRDCJT9t

[12] Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. p. 7

Where the paper says it

[12] Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow.

arXiv preprint arXiv:2209. p. 7
Where the paper says it

Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.

03003, 2022. p. 7
Where the paper says it

arXiv preprint arXiv:2209. 03003, 2022.

[13] Junzhe Lu, Jing Lin, Hongkun Dou, Ailing Zeng, Yue Deng, Yulun Zhang, and Haoqian Wang. p. 7

Where the paper says it

03003, 2022. [13] Junzhe Lu, Jing Lin, Hongkun Dou, Ailing Zeng, Yue Deng, Yulun Zhang, and Haoqian Wang.

Dposer: Diffusion model as robust 3d human pose prior, 2024. p. 7
Where the paper says it

[13] Junzhe Lu, Jing Lin, Hongkun Dou, Ailing Zeng, Yue Deng, Yulun Zhang, and Haoqian Wang. Dposer: Diffusion model as robust 3d human pose prior, 2024.

URL https://arxiv.org/abs/2312. p. 7
Where the paper says it

URL https://arxiv. org/abs/2312.

05541

[14] Junzhe Lu, Jing Lin, Hongkun Dou, Ailing Zeng, Yue Deng, Xian Liu, Zhongang Cai, Lei Yang, Yulun Zhang, Haoqian Wang, and Ziwei Liu. Dposer-x: Diffusion model as robust 3d whole-body human pose prior. p. 7

Where the paper says it

[14] Junzhe Lu, Jing Lin, Hongkun Dou, Ailing Zeng, Yue Deng, Xian Liu, Zhongang Cai, Lei Yang, Yulun Zhang, Haoqian Wang, and Ziwei Liu. Dposer-x: Diffusion model as robust 3d whole-body human pose prior.

arXiv preprint arXiv:2508. p. 7
Where the paper says it

Dposer-x: Diffusion model as robust 3d whole-body human pose prior. arXiv preprint arXiv:2508.

00599

[15] Zhengyi Luo, Jinkun Cao, Kris Kitani, Weipeng Xu, et al. Perpetual humanoid control for real-time simulated avatars. p. 7

Where the paper says it

[15] Zhengyi Luo, Jinkun Cao, Kris Kitani, Weipeng Xu, et al. Perpetual humanoid control for real-time simulated avatars.

In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 10895–10904, 2023. p. 7
Where the paper says it

Perpetual humanoid control for real-time simulated avatars. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 10895–10904, 2023.

[16] Zhengyi Luo, Jinkun Cao, Josh Merel, Alexander Winkler, Jing Huang, Kris Kitani, and Weipeng Xu. p. 7

Where the paper says it

In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 10895–10904, 2023. [16] Zhengyi Luo, Jinkun Cao, Josh Merel, Alexander Winkler, Jing Huang, Kris Kitani, and Weipeng Xu.

Universal humanoid motion representations for physics-based control. p. 7
Where the paper says it

[16] Zhengyi Luo, Jinkun Cao, Josh Merel, Alexander Winkler, Jing Huang, Kris Kitani, and Weipeng Xu. Universal humanoid motion representations for physics-based control.

In International Conference on Learning Representations, volume 2024, pages 56766–56782, 2024. p. 7
Where the paper says it

Universal humanoid motion representations for physics-based control. In International Conference on Learning Representations, volume 2024, pages 56766–56782, 2024.

[17] Zhengyi Luo, Ye Yuan, Tingwu Wang, Chenran Li, Sirui Chen, Fernando Castaneda, Zi-Ang Cao, Jiefeng˜ Li, David Minor, Qingwei Ben, Xingye Da, Runyu Ding. p. 7

Where the paper says it

In International Conference on Learning Representations, volume 2024, pages 56766–56782, 2024. [17] Zhengyi Luo, Ye Yuan, Tingwu Wang, Chenran Li, Sirui Chen, Fernando Castaneda, Zi-Ang Cao, Jiefeng ˜ Li, David Minor, Qingwei Ben, Xingye Da, Runyu Ding, Cyrus Hogg, Lina Song, Edy Lim, Eugene Jeong, Tairan He, Haoru Xue, Wenli Xiao, Zi Wang, Simon Yuen, Jan Kautz, Yan Chang, Umar Iqbal, Linxi Fan, and Yuke Zhu.

Sonic: Supersizing motion tracking for natural humanoid whole-body control. p. 7
Where the paper says it

[17] Zhengyi Luo, Ye Yuan, Tingwu Wang, Chenran Li, Sirui Chen, Fernando Castaneda, Zi-Ang Cao, Jiefeng ˜ Li, David Minor, Qingwei Ben, Xingye Da, Runyu Ding, Cyrus Hogg, Lina Song, Edy Lim, Eugene Jeong, Tairan He, Haoru Xue, Wenli Xiao, Zi Wang, Simon Yuen, Jan Kautz, Yan Chang, Umar Iqbal, Linxi Fan, and Yuke Zhu. Sonic: Supersizing motion tracking for natural humanoid whole-body control.

arXiv preprint arXiv:2511.07820

[18] Naureen Mahmood, Nima Ghorbani, Nikolaus F Troje, Gerard Pons-Moll, and Michael J Black. Amass: Archive of motion capture as surface shapes. p. 7

Where the paper says it

[18] Naureen Mahmood, Nima Ghorbani, Nikolaus F Troje, Gerard Pons-Moll, and Michael J Black. Amass: Archive of motion capture as surface shapes.

In Proceedings of the IEEE/CVF international conference on computer vision, pages 5442–5451, 2019. p. 7
Where the paper says it

Amass: Archive of motion capture as surface shapes. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5442–5451, 2019.

[19] Hila Manor and Tomer Michaeli. p. 7

Where the paper says it

In Proceedings of the IEEE/CVF international conference on computer vision, pages 5442–5451, 2019. [19] Hila Manor and Tomer Michaeli.

On the posterior distribution in denoising: Application to uncertainty quantification. p. 7
Where the paper says it

[19] Hila Manor and Tomer Michaeli. On the posterior distribution in denoising: Application to uncertainty quantification.

arXiv preprint arXiv:2309. p. 7
Where the paper says it

On the posterior distribution in denoising: Application to uncertainty quantification. arXiv preprint arXiv:2309.

13598, 2023. p. 7
Where the paper says it

arXiv preprint arXiv:2309. 13598, 2023.

[20] Yuxuan Mu, Ziyu Zhang, Yi Shi, Dun Yang, Minami Matsumoto, Kotaro Imamura, Guy Tevet, Chuan Guo, Michael Taylor, Chang Shu, Pengcheng Xi, and Xue Bin Peng. p. 7

Where the paper says it

13598, 2023. [20] Yuxuan Mu, Ziyu Zhang, Yi Shi, Dun Yang, Minami Matsumoto, Kotaro Imamura, Guy Tevet, Chuan Guo, Michael Taylor, Chang Shu, Pengcheng Xi, and Xue Bin Peng.

Smp: Reusable score-matching motion priors for physics-based character control. p. 7
Where the paper says it

[20] Yuxuan Mu, Ziyu Zhang, Yi Shi, Dun Yang, Minami Matsumoto, Kotaro Imamura, Guy Tevet, Chuan Guo, Michael Taylor, Chang Shu, Pengcheng Xi, and Xue Bin Peng. Smp: Reusable score-matching motion priors for physics-based character control.

ACM Transactions on Graphics (Proceedings of SIGGRAPH 2026), 2026. p. 7
Where the paper says it

Smp: Reusable score-matching motion priors for physics-based character control. ACM Transactions on Graphics (Proceedings of SIGGRAPH 2026), 2026.

[21] Jebastin Nadar, Simone Foti, and Tolga Birdal. p. 7

Where the paper says it

ACM Transactions on Graphics (Proceedings of SIGGRAPH 2026), 2026. [21] Jebastin Nadar, Simone Foti, and Tolga Birdal.

Posedflow: Versatile and guided flow matching model of human pose. p. 7
Where the paper says it

[21] Jebastin Nadar, Simone Foti, and Tolga Birdal. Posedflow: Versatile and guided flow matching model of human pose.

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21165–21175, 2026. p. 7
Where the paper says it

Posedflow: Versatile and guided flow matching model of human pose. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21165–21175, 2026.

[22] William Peebles and Saining Xie. p. 7

Where the paper says it

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21165–21175, 2026. [22] William Peebles and Saining Xie.

Scalable diffusion models with transformers. p. 7
Where the paper says it

[22] William Peebles and Saining Xie. Scalable diffusion models with transformers.

In Proceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023. p. 7
Where the paper says it

Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023.

[23] Xue Bin Peng. p. 7

Where the paper says it

Accessed: 2026-05-06. [3] Zixuan Chen, Mazeyu Ji, Xuxin Cheng, Xuanbin Peng, Xue Bin Peng, and Xiaolong Wang.

Mimickit: A reinforcement learning framework for motion imitation and control. p. 7
Where the paper says it

[23] Xue Bin Peng. Mimickit: A reinforcement learning framework for motion imitation and control.

arXiv preprint arXiv:2510. p. 7
Where the paper says it

Mimickit: A reinforcement learning framework for motion imitation and control. arXiv preprint arXiv:2510.

13794, 2025. p. 7
Where the paper says it

arXiv preprint arXiv:2510. 13794, 2025.

[24] Xue Bin Peng, Pieter Abbeel, Sergey Levine, and Michiel Van de Panne. p. 7

Where the paper says it

13794, 2025. [24] Xue Bin Peng, Pieter Abbeel, Sergey Levine, and Michiel Van de Panne.

Deepmimic: Example-guided deep re-

[25] Xue Bin Peng, Ze Ma, Pieter Abbeel, Sergey Levine, and Angjoo Kanazawa. Amp: Adversarial motion priors for stylized physics-based character control. p. 8

Where the paper says it

[25] Xue Bin Peng, Ze Ma, Pieter Abbeel, Sergey Levine, and Angjoo Kanazawa. Amp: Adversarial motion priors for stylized physics-based character control.

ACM Transactions on Graphics (ToG), 40(4):1–20, 2021. p. 8
Where the paper says it

Amp: Adversarial motion priors for stylized physics-based character control. ACM Transactions on Graphics (ToG), 40(4):1–20, 2021.

[26] Xue Bin Peng, Yunrong Guo, Lina Halper, Sergey Levine, and Sanja Fidler. p. 8

Where the paper says it

ACM Transactions on Graphics (ToG), 40(4):1–20, 2021. [26] Xue Bin Peng, Yunrong Guo, Lina Halper, Sergey Levine, and Sanja Fidler.

Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters. p. 8
Where the paper says it

[26] Xue Bin Peng, Yunrong Guo, Lina Halper, Sergey Levine, and Sanja Fidler. Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters.

ACM Transactions On Graphics (TOG), 41(4):1–17, 2022. p. 8
Where the paper says it

Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters. ACM Transactions On Graphics (TOG), 41 (4):1–17, 2022.

[27] Zelin Tao, Zeran Su, Peiran Liu, Jingkai Sun, Wenqiang Que, Jiahao Ma, Jialin Yu, Jiahang Cao, Pihai Sun, Hao Liang, et al. p. 8

Where the paper says it

ACM Transactions On Graphics (TOG), 41 (4):1–17, 2022. [27] Zelin Tao, Zeran Su, Peiran Liu, Jingkai Sun, Wenqiang Que, Jiahao Ma, Jialin Yu, Jiahang Cao, Pihai Sun, Hao Liang, et al.

Heracles: Bridging precise tracking and generative synthesis for general humanoid control. p. 8
Where the paper says it

[27] Zelin Tao, Zeran Su, Peiran Liu, Jingkai Sun, Wenqiang Que, Jiahao Ma, Jialin Yu, Jiahang Cao, Pihai Sun, Hao Liang, et al. Heracles: Bridging precise tracking and generative synthesis for general humanoid control.

arXiv preprint arXiv:2603.27756

[28] Chen Tessler, Yoni Kasten, Yunrong Guo, Shie Mannor, Gal Chechik, and Xue Bin Peng. Calm: Conditional adversarial latent models for directable virtual characters. p. 8

Where the paper says it

[28] Chen Tessler, Yoni Kasten, Yunrong Guo, Shie Mannor, Gal Chechik, and Xue Bin Peng. Calm: Conditional adversarial latent models for directable virtual characters.

In ACM SIGGRAPH 2023 conference proceedings, pages 1–9, 2023. p. 8
Where the paper says it

Calm: Conditional adversarial latent models for directable virtual characters. In ACM SIGGRAPH 2023 conference proceedings, pages 1–9, 2023.

[29] Chen Tessler, Yunrong Guo, Ofir Nabati, Gal Chechik, and Xue Bin Peng. p. 8

Where the paper says it

In ACM SIGGRAPH 2023 conference proceedings, pages 1–9, 2023. [29] Chen Tessler, Yunrong Guo, Ofir Nabati, Gal Chechik, and Xue Bin Peng.

Maskedmimic: Unified physics-based character control through masked motion inpainting. p. 8
Where the paper says it

[29] Chen Tessler, Yunrong Guo, Ofir Nabati, Gal Chechik, and Xue Bin Peng. Maskedmimic: Unified physics-based character control through masked motion inpainting.

ACM Transactions on Graphics (TOG), 2024. p. 8
Where the paper says it

Maskedmimic: Unified physics-based character control through masked motion inpainting. ACM Transactions on Graphics (TOG), 2024.

[30] Garvita Tiwari, Dimitrije Antic, Jan Eric Lenssen, Niko-laos Sarafianos, Tony Tung, and Gerard Pons-Moll. p. 8

Where the paper says it

ACM Transactions on Graphics (TOG), 2024. [30] Garvita Tiwari, Dimitrije Antic, Jan Eric Lenssen, Niko- ´ laos Sarafianos, Tony Tung, and Gerard Pons-Moll.

Posendf: Modeling human pose manifolds with neural distance fields. p. 8
Where the paper says it

[30] Garvita Tiwari, Dimitrije Antic, Jan Eric Lenssen, Niko- ´ laos Sarafianos, Tony Tung, and Gerard Pons-Moll. Posendf: Modeling human pose manifolds with neural distance fields.

In European Conference on Computer Vision, pages 572–589. p. 8
Where the paper says it

Posendf: Modeling human pose manifolds with neural distance fields. In European Conference on Computer Vision, pages 572–589.

Springer, 2022. p. 8
Where the paper says it

In European Conference on Computer Vision, pages 572–589. Springer, 2022.

[31] Jungdam Won, Deepak Gopinath, and Jessica Hodgins. p. 8

Where the paper says it

Springer, 2022. [31] Jungdam Won, Deepak Gopinath, and Jessica Hodgins.

Physics-based character controllers using conditional vaes. p. 8
Where the paper says it

[31] Jungdam Won, Deepak Gopinath, and Jessica Hodgins. Physics-based character controllers using conditional vaes.

ACM Trans. Graph., 41(4), 2022. p. 7
Where the paper says it

Adversarial motion priors make good substitutes for complex reward functions. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 25–32.

[32] Xiaofeng Yang, Cheng Chen, Fayao Liu, Guosheng Lin, et al. p. 8

Where the paper says it

1145/3528223. [32] Xiaofeng Yang, Cheng Chen, Fayao Liu, Guosheng Lin, et al.

Text-to-image rectified flow as plug-and-play priors. p. 8
Where the paper says it

[32] Xiaofeng Yang, Cheng Chen, Fayao Liu, Guosheng Lin, et al. Text-to-image rectified flow as plug-and-play priors.

In International Conference on Learning Representations, volume 2025, pages 13896–13920, 2025. p. 8
Where the paper says it

Text-to-image rectified flow as plug-and-play priors. In International Conference on Learning Representations, volume 2025, pages 13896–13920, 2025.

[33] Heyuan Yao, Zhenhua Song, Baoquan Chen, and Libin Liu. p. 8

Where the paper says it

In International Conference on Learning Representations, volume 2025, pages 13896–13920, 2025. [33] Heyuan Yao, Zhenhua Song, Baoquan Chen, and Libin Liu.

ControlVAE: Model-Based Learning of Generative Controllers for Physics-Based Characters. p. 8
Where the paper says it

[33] Heyuan Yao, Zhenhua Song, Baoquan Chen, and Libin Liu. ControlVAE: Model-Based Learning of Generative Controllers for Physics-Based Characters.

ACM Transactions on Graphics, 41(6):183:1–183:16, 2022. p. 8
Where the paper says it

ControlVAE: Model-Based Learning of Generative Controllers for Physics-Based Characters. ACM Transactions on Graphics, 41(6):183:1–183:16, 2022.

[34] Heyuan Yao, Zhenhua Song, Yuyang Zhou, Tenglong Ao, Baoquan Chen, and Libin Liu. p. 8

Where the paper says it

ISSN 0730-0301. [34] Heyuan Yao, Zhenhua Song, Yuyang Zhou, Tenglong Ao, Baoquan Chen, and Libin Liu.

Moconvq: Unified physics-based motion control via scalable discrete representations. p. 8
Where the paper says it

[34] Heyuan Yao, Zhenhua Song, Yuyang Zhou, Tenglong Ao, Baoquan Chen, and Libin Liu. Moconvq: Unified physics-based motion control via scalable discrete representations.

ACM Transactions on Graphics (TOG), 43(4):1–21, 2024

[35] Shaofeng Yin, Yanjie Ze, Hong-Xing Yu, C. Karen Liu, and Jiajun Wu. p. 8

Where the paper says it

[35] Shaofeng Yin, Yanjie Ze, Hong-Xing Yu, C. Karen Liu, and Jiajun Wu.

Visualmimic: Visual humanoid locomanipulation via motion tracking and generation. p. 8
Where the paper says it

Karen Liu, and Jiajun Wu. Visualmimic: Visual humanoid locomanipulation via motion tracking and generation.

arXiv preprint arXiv:2509. p. 8
Where the paper says it

Visualmimic: Visual humanoid locomanipulation via motion tracking and generation. arXiv preprint arXiv:2509.

p. 8
Where the paper says it

Visualmimic: Visual humanoid locomanipulation via motion tracking and generation. arXiv preprint arXiv:2509.

p. 8
Where the paper says it

Visualmimic: Visual humanoid locomanipulation via motion tracking and generation. arXiv preprint arXiv:2509.

p. 8
Where the paper says it

Visualmimic: Visual humanoid locomanipulation via motion tracking and generation. arXiv preprint arXiv:2509.

20322

[36] Zhengdi Yu, Simone Foti, Linguang Zhang, Amy Zhao, Cem Keskin, Stefanos Zafeiriou, and Tolga Birdal. Geometric neural distance fields for learning human motion priors. p. 8

Where the paper says it

[36] Zhengdi Yu, Simone Foti, Linguang Zhang, Amy Zhao, Cem Keskin, Stefanos Zafeiriou, and Tolga Birdal. Geometric neural distance fields for learning human motion priors.

arXiv preprint arXiv:2509.09667

[37] Siwei Zhang, Bharat Lal Bhatnagar, Yuanlu Xu, Alexander Winkler, Petr Kadlecek, Siyu Tang, and Federica Bogo. Rohm: Robust human motion reconstruction via diffusion. p. 8

Where the paper says it

[37] Siwei Zhang, Bharat Lal Bhatnagar, Yuanlu Xu, Alexander Winkler, Petr Kadlecek, Siyu Tang, and Federica Bogo. Rohm: Robust human motion reconstruction via diffusion.

In CVPR, 2024. p. 7
Where the paper says it

Nrdf: Neural riemannian distance fields for learning articulated pose priors. In Conference on Computer Vision and Pattern Recognition (CVPR), 2024.

[38] Ziyu Zhang, Sergey Bashkirov, Dun Yang, Yi Shi, Michael Taylor, and Xue Bin Peng. p. 8

Where the paper says it

In CVPR, 2024. [38] Ziyu Zhang, Sergey Bashkirov, Dun Yang, Yi Shi, Michael Taylor, and Xue Bin Peng.

Physics-based motion imitation with adversarial differential discriminators. p. 8
Where the paper says it

[38] Ziyu Zhang, Sergey Bashkirov, Dun Yang, Yi Shi, Michael Taylor, and Xue Bin Peng. Physics-based motion imitation with adversarial differential discriminators.

In Proceedings of the SIGGRAPH Asia 2025 Conference Papers, pages 1–12, 2025. p. 8
Where the paper says it

Physics-based motion imitation with adversarial differential discriminators. In Proceedings of the SIGGRAPH Asia 2025 Conference Papers, pages 1–12, 2025.

[39] Siheng Zhao, Yanjie Ze, Yue Wang, C. p. 8

Where the paper says it

In Proceedings of the SIGGRAPH Asia 2025 Conference Papers, pages 1–12, 2025. [39] Siheng Zhao, Yanjie Ze, Yue Wang, C.

Karen Liu, Pieter Abbeel, Guanya Shi, and Rocky Duan. p. 8
Where the paper says it

[39] Siheng Zhao, Yanjie Ze, Yue Wang, C. Karen Liu, Pieter Abbeel, Guanya Shi, and Rocky Duan.

Resmimic: From general motion tracking to humanoid whole-body locomanipulation via residual learning, 2025. p. 8
Where the paper says it

Karen Liu, Pieter Abbeel, Guanya Shi, and Rocky Duan. Resmimic: From general motion tracking to humanoid whole-body locomanipulation via residual learning, 2025.

URL https://arxiv.org/abs/2510. p. 8
Where the paper says it

URL https: //arxiv. org/abs/2510.

05070

[40] Mingyuan Zhou, Yi Gu, Huangjie Zheng, Liangchen Song, Guande He, Yizhe Zhang, Wenze Hu, and Yinfei Yang. Score distillation of flow matching models. p. 8

Where the paper says it

[40] Mingyuan Zhou, Yi Gu, Huangjie Zheng, Liangchen Song, Guande He, Yizhe Zhang, Wenze Hu, and Yinfei Yang. Score distillation of flow matching models.

p. 8
Where the paper says it

[40] Mingyuan Zhou, Yi Gu, Huangjie Zheng, Liangchen Song, Guande He, Yizhe Zhang, Wenze Hu, and Yinfei Yang. Score distillation of flow matching models.

arXiv preprint arXiv:2509.25127

[41] Qingxu Zhu, He Zhang, Mengting Lan, and Lei Han. Neural categorical priors for physics-based character control. p. 8

Where the paper says it

[41] Qingxu Zhu, He Zhang, Mengting Lan, and Lei Han. Neural categorical priors for physics-based character control.

p. 8
Where the paper says it

[41] Qingxu Zhu, He Zhang, Mengting Lan, and Lei Han. Neural categorical priors for physics-based character control.

ACM Transactions on Graphics (TOG), 42(6): 1–16, 2023

[40] Mingyuan Zhou, Yi Gu, Huangjie Zheng, Liangchen Song, Guande He, Yizhe Zhang, Wenze Hu, and Yinfei Yang. Score distillation of flow matching models. arXiv preprint arXiv:2509.25127

[41] Qingxu Zhu, He Zhang, Mengting Lan, and Lei Han. Neural categorical priors for physics-based character control. ACM Transactions on Graphics (TOG), 42(6): 1–16, 2023

Supplementary Material Details

The population-optimal PGS is large when the stabilized rollout direction has substantial components along posterior pose-variation modes with large conditional covariance eigenvalues The authors use a single JVP as a stable and efficient online PGS estimator during policy training. PFM-HR is trained with a residual MLP with 10 blocks and a hidden dimension of 1024. p. 4

Where the paper says it

All policies are trained on 8 NVIDIA RTX 4090 GPUs using 4096 parallel environments and the default MimicKit configuration. PFM-HR uses a residual MLP with 10 blocks and a hidden dimension of 1024.

The latency for one online PGS evaluation step using a batch size of 4096 on an NVIDIA GeForce RTX 4090 GPU is 0. p. 5
Where the paper says it

With a batch size of 4096 on an NVIDIA GeForce RTX 4090 GPU, three denoiser evaluations require 1. 8 ms, whereas one Jacobian– vector product requires 0.

75 ms. PFM-HR requires the fewest simulation samples to reach SR ≥ 80% across all skills, reducing the requirements by 15. p. 4
Where the paper says it

In Table I, BM denotes the original BeyondMimic method without an auxiliary pose prior, while BM w/ PDF-HR and BM w/ PFM-HR augment BM with PDFHR and our PFM-HR, respectively. BM w/ PFM-HR requires the fewest simulation samples to reach SR ≥ 80% across all skills, reducing the requirements by 15.

1%–30.8% relative to BM. p. 0
Where the paper says it

Recent frozen priors improve modularity but introduce a different trade-off. PDF-HR learns a reusable pose-distance field from retargeted humanoid data and improves downstream tracking [5], but it requires precomputed pose–distance supervision and scores each pose without considering the relative joint changes that produced it.

The best means are highlighted in Table I, and PFM-HR achieves the lowest or tied-lowest positional error on eight tasks. p. 4
Where the paper says it

ADD w/ PFM-HR also achieves the lowest or tied-lowest positional error on eight tasks, although the margins are generally small. Overall, PFM-HR provides the best balance between convergence and tracking accuracy, particularly on dynamic motions.

The average discrimination score increases from 0. p. 3
Where the paper says it

A high PGS therefore indicates a joint-change pattern strongly supported by the pose prior. After calibration, PGS increases the reward contribution of such coordinated updates and reduces that of weakly supported patterns, biasing policy optimization toward coordinated humanoid poses.

043 at teval = 0. p. 2
Where the paper says it

(9) We denote the learned approximation to this conditional mean by Fϕ(z, t) := xˆϕ(z, t). (10) During policy learning, Fϕ remains frozen and is queried at (qk−1 , teval), where teval ≤ 1 − δ.

4 to 0.124 at teval = 0. p. 2
Where the paper says it

(9) We denote the learned approximation to this conditional mean by Fϕ(z, t) := xˆϕ(z, t). (10) During policy learning, Fϕ remains frozen and is queried at (qk−1 , teval), where teval ≤ 1 − δ.

75, and subsequently decreases to 0.075 at teval = 0. p. 2
Where the paper says it

(9) We denote the learned approximation to this conditional mean by Fϕ(z, t) := xˆϕ(z, t). (10) During policy learning, Fϕ remains frozen and is queried at (qk−1 , teval), where teval ≤ 1 − δ.

9. The neighboring timesteps 0. p. 3
Where the paper says it

Each policy is trained for 12 billion samples and evaluated with tracking horizons of 10, 20, and 30 seconds, while all other settings are held fixed. Figure 3 shows that intermediate timesteps provide faster convergence and lower position errors.

7 and 0.8 also obtain relatively strong average scores of 0. p. 3
Where the paper says it

Figure 3 shows that intermediate timesteps provide faster convergence and lower position errors. Strong corruption at small t reduces pose specificity, whereas near t = 1 the denoising map approaches the identity and becomes less discriminative.

116 and 0.119, indicating that PGS remains effective within an intermediate timestep range rather than depending on a narrowly tuned value. A larger positive value indicates that PGS assigns substantially lower reference-relative scores after the joint-change structure is disrupted and therefore provides stronger structural discrimination. p. 0
Where the paper says it

A high PGS indicates that the relative joint changes align with co-variation modes strongly represented by the pose prior. After reference-specific calibration, PGS modulates the tracking reward to reduce learning signals from weakly supported change patterns.

The normalized score remains within 1. p. 2
Where the paper says it

(10) During policy learning, Fϕ remains frozen and is queried at (qk−1 , teval), where teval ≤ 1 − δ. Pose Geometry Score The Pose Geometry Score measures whether the relative joint changes produced by the policy agree with the conditional co-variation structure encoded by the pose prior.

6% of the unperturbed value, and the calibration-region agreement remains above 99%. p. 2
Where the paper says it

The policy is trained to maximize J(πθ) = Eτ∼πθ " T X−1 k=0 γ k r(sk, ak) # . (3) In humanoid motion tracking, the reward commonly combines task and tracking terms, rk = w Gr G k + w T r T k , (4) where r G k measures task progress, r T k measures agreement with the reference motion, and w G and w T are scalar weights.

In contrast, perturbations that progressively disrupt the inter-joint coordination pattern are expected to reduce RPGS, decrease the calibrated percentile and reward multiplier. The results show that PGS is robust to nuisance variations while remaining sensitive to the inter-j… p. 6
Where the paper says it

The reliability of PGS also depends on the coverage of the pose dataset, and its guidance may be less informative for poses far from the training distribution. Future work will explore sign-sensitive scores and temporally conditioned priors while preserving the efficiency and reusability of PFM-HR.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

1

Improve humanoid locomotion control by guiding policy exploration toward structured pose changes using PGS to modulate tracking rewards. This guides policy optimization toward structured pose changes while keeping the prior frozen across tracking tasks.

2

Enhance sample efficiency and tracking accuracy for dynamic and acrobatic motions by leveraging PFM-HR's ability to provide geometry-aware guidance. The paper demonstrates that PFM-HR provides the best balance between convergence and tracking accuracy, particularly on dynamic motions.

3

Improve long-horizon general motion tracking reliability by using PGS to reduce position error by 7.6% over ADD and 10.3% over PDF-HR across different episode lengths. This shows that PGS improves both learning speed and long-horizon tracking reliability rather than trading one off against the other.

4

Enable real-world humanoid deployment compatibility by using PFM-HR as a frozen prior during simulation training without requiring additional deployment time computation. This allows policies trained with PFM-HR to execute diverse highly dynamic whole-body motions on a humanoid robot.

Abstract

Motion priors improve reinforcement learning for physics-based humanoid tracking, but temporal priors require ordered motion clips, while pose priors provide limited guidance for policy-induced pose transitions. We present Pose Flow Matching for Humanoid Robots (PFM-HR), a reusable flow matching prior trained directly on large scale unordered pose data. PFM-HR introduces the Pose Geometry Score (PGS), which quantifies how joint coordinate changes during rollouts align with the local geometry of pose variation captured by the prior. Using PGS to modulate the tracking reward guides policy exploration toward structured pose changes while keeping the prior frozen across tracking tasks. Experiments demonstrate that PFM-HR improves both single motion and general motion tracking, especially for highly dynamic motions.

Sources

Related papers