Paper Sweeps

Humanoid Papers from October 11 to October 17, 2026

8Locomotion14Motion Tracking12Loco-Manipulation11Whole-Body Control21Training Infrastructure8State Estimation / Perception

8 papers

Locomotion

Terrain, gait, robustness, and agile movement.

paper preview Locomotion with Online Awareness of Dynamics for Precision Footholds visual

01arXiv

Locomotion with Online Awareness of Dynamics for Precision Footholds

Helps quadrupeds place their feet precisely by estimating how payloads change their dynamics online.

Kunzhao Ren et al.

project clip

02arXiv

P-BFM: Perceptive Behavioral Foundation Models for Humanoid Control via Unsupervised Reinforcement Learning

Gives one humanoid controller a library of behaviors that adapt to terrain and steering commands.

Xue Wang et al.

paper preview Geometry-Aware Predictive Control for Walk-to-Run Bipedal Obstacle Crossing visual

03arXiv

Geometry-Aware Predictive Control for Walk-to-Run Bipedal Obstacle Crossing

Plans a biped's transition from walking to running so it can clear obstacles without losing balance.

Aashish Sahu et al.

paper preview Push the Limit: Learning High-Speed Sprinting and Aggressive Cornering on a Full-Scale Humanoid visual

04arXiv

Push the Limit: Learning High-Speed Sprinting and Aggressive Cornering on a Full-Scale Humanoid

Teaches a full-size humanoid to sprint quickly and corner aggressively without falling.

Huaxing Huang et al.

project clip

05arXiv

Terrain Traversal Without Motion Priors: Expert-Routed Humanoid Locomotion at Deployment Fidelity

Routes a humanoid among specialized walking experts to cross varied terrain without motion demonstrations.

Thanh Ly et al.

paper preview Design and Analysis of a Nonlinear Parallel Mechanism Coupling for Dynamic Humanoid Legs visual

06arXiv

Design and Analysis of a Nonlinear Parallel Mechanism Coupling for Dynamic Humanoid Legs

Shares actuator effort across a humanoid leg mechanism to improve squatting and dynamic motion.

Rohit Kumar et al.

paper preview PolyMorph: Learning Locomotion Policy for Cross-Embodiment Adaptation visual

07arXiv

PolyMorph: Learning Locomotion Policy for Cross-Embodiment Adaptation

Trains one locomotion policy that adapts across bipeds, tripeds, and quadrupeds.

Hyunsik Oh et al.

paper preview Persistent Profile-Guided Reset-Free Recovery Improves Final Gait Selection after Irreversible Leg Damage visual

08arXiv

Persistent Profile-Guided Reset-Free Recovery Improves Final Gait Selection after Irreversible Leg Damage

Helps a damaged legged robot choose a stable new gait without repeated resets.

Minjoong Kim

14 papers

Motion Tracking

Reference motion, imitation, retargeting, and behavior priors.

paper preview PartiGen: Part-Aware Representations for Language-Conditioned Humanoid Motion Generation visual

09arXiv

PartiGen: Part-Aware Representations for Language-Conditioned Humanoid Motion Generation

Generates whole-body G1 motions from language while keeping each body part coordinated.

Chuxuan Huang et al.

paper preview PriMoRetarget: Humanoid Motion Transfer Through Prioritized Motion Retargeting visual

10arXiv

PriMoRetarget: Humanoid Motion Transfer Through Prioritized Motion Retargeting

Retargets human motions to humanoids while preserving foot contacts and interaction constraints.

Nicholas Ioannidis et al.

project clip

11arXiv

Sonic-AMG: Streaming Human Motion Generation from Spatial Audio

Generates humanoid motions that react continuously to the direction and timing of sounds.

Jingcheng Ni et al.

paper preview DirtyMoCap: Robust Motion Capture from Unconstrained Markers visual

12arXiv

DirtyMoCap: Robust Motion Capture from Unconstrained Markers

Recovers human motion from loose, noisy optical markers so it can become reusable robot-reference data.

Long Wang et al.

project clip

13arXiv

PAMI: Part Anchored Motion for Text to Human-Object Interaction Generation

Generates people interacting with objects from text while maintaining body-object contact.

Chuqiao Li et al.

project clip

14arXiv

EgoExplorer: A Vision-Motion-Touch Observatory for Human Loco-Manipulation In Situ

Captures vision, body motion, hand pose, and touch together for humanoid loco-manipulation demonstrations.

Yunsong Zhou et al.

paper preview Walk Like This: Stylized Locomotion for Physics-Based Characters through Example-Driven Diffusion visual

15arXiv

Walk Like This: Stylized Locomotion for Physics-Based Characters through Example-Driven Diffusion

Generates stylized G1 walking and turning from a few example motions.

Joshua Aurand et al.

project clip

16arXiv

TRACE: Efficient Transfer Learning for Neural Whole-Body Control Across Humanoid Embodiments

Transfers a trained whole-body controller from Unitree G1 to H2 with a small amount of adaptation.

Qingyu Lan et al.

paper preview CopyQuad: Quadrupedal Skill Transfer from a Single Side-View Video Without a Depth Model visual

17arXiv

CopyQuad: Quadrupedal Skill Transfer from a Single Side-View Video Without a Depth Model

Learns quadruped tricks from a single side-view video without estimating full 3D motion.

Mohamad H. Danesh et al.

paper preview Embodiment-Aware Inference-Time Search for Human Reaction Selection in Humanoid Retargeting visual

18arXiv

Embodiment-Aware Inference-Time Search for Human Reaction Selection in Humanoid Retargeting

Chooses human reaction motions that remain physically feasible after retargeting to a humanoid.

Yongpeng Cao et al.

project clip

19arXiv

KinoGen: Customizable Humanoid Loco-Manipulation References via Kinematically Grounded Video Generation

Generates editable humanoid and object motion references from video for G1 whole-body tracking.

Qi Wu et al.

paper preview Beyond Kinematic Tracking: Preserving Expressive Characteristics in Humanoid Whole-Body Control visual

20arXiv

Beyond Kinematic Tracking: Preserving Expressive Characteristics in Humanoid Whole-Body Control

Restores the energy, posture, and style that standard humanoid motion trackers often lose.

Lingfan Bao et al.

project visual Compliant GMT: Making General Motion Tracking Compliant via Residual Reference Adaptation visual

21Project

Compliant GMT: Making General Motion Tracking Compliant via Residual Reference Adaptation

Makes a general humanoid motion tracker compliant by adapting references during physical contact.

Venki Parsuram et al.

paper preview TRACE: Typed Reference-Aware Conversational Editing for Humanoid Motion Diffusion visual

22arXiv

TRACE: Typed Reference-Aware Conversational Editing for Humanoid Motion Diffusion

Edits generated humanoid motions with natural language while preserving protected body constraints.

Zejian Kang et al.

12 papers

Loco-Manipulation

Whole-body task contact: feet, torso, hands, and objects.

project clip

23arXiv

ATHLETE: Learning Reactive Humanoid Tennis via Trajectory-Guided Motion Matching

Teaches humanoids to play reactive tennis by combining walking skills with video-learned swings.

Zhihao Tian et al.

project clip

24arXiv

Belief-Overlap Partial Inversion: Object-Centric Flow Replanning for Humanoid Loco-Manipulation

Replans a humanoid's cart-pushing motion when the cart's mass or caster behavior changes.

Chen Yu et al.

paper preview VICAR: Video Imitation with Contact-Aware Retargeting for Humanoid Interactive Skills visual

25arXiv

VICAR: Video Imitation with Contact-Aware Retargeting for Humanoid Interactive Skills

Turns ordinary videos into contact-aware humanoid skills such as climbing a ladder or opening objects.

Dvij Kalaria et al.

project clip

26arXiv

PanTact: Teaching Whole-Body Dexterity with a Wearable Contact Interface

Teaches a humanoid to step, brace, and carry with its whole body through a wearable contact interface.

Xiaomeng Xu et al.

paper preview Learning Safety-Constrained Whole-Body Loco-Manipulation for Lunar Surface Operations visual

27arXiv

Learning Safety-Constrained Whole-Body Loco-Manipulation for Lunar Surface Operations

Teaches a humanoid to move its legs and arms safely while working in low-gravity lunar conditions.

Alvaro Belmonte-Baeza et al.

paper preview From Video to Vision: Learning Generalizable Humanoid Carrying from Human Motion visual

28arXiv

From Video to Vision: Learning Generalizable Humanoid Carrying from Human Motion

Teaches humanoids to carry objects from human video while adapting to new views and loads.

Ziluo Ding

paper preview Composable Dynamics for Cooperative Humanoids: A RoboCompiler Extension visual

29arXiv

Composable Dynamics for Cooperative Humanoids: A RoboCompiler Extension

Compiles cooperative humanoid tasks into coordinated behaviors shared across multiple robots.

Mehdi Heydari Shahna et al.

project clip

30arXiv

From Hitter to Player: Separating Motor Execution and Strategy in Competitive Humanoid Table Tennis

Separates footwork and striking from game strategy so a humanoid can play competitive table tennis.

Fanrong Dong et al.

paper preview OIL: An Object Interaction Layer for Humanoid Whole-Body Control visual

31arXiv

OIL: An Object Interaction Layer for Humanoid Whole-Body Control

Adds a reusable object-interaction layer to whole-body humanoid controllers.

Pranay Dugar et al.

project clip

32arXiv

PingpongDuo: Agile Humanoid Table Tennis Doubles with Safe Team Coordination

Coordinates two humanoids so they can move, take turns, and return table-tennis shots safely.

Haoran Jiang et al.

paper preview Whole-Body Dexterity visual

33arXiv

Whole-Body Dexterity

Lets a teleoperated humanoid slide and hand off objects across its arms and torso while balancing.

Conor Mc Gartoll et al.

paper preview SOMA: Stack-of-Objects Manipulation using a Quadruped Robot visual

34arXiv

SOMA: Stack-of-Objects Manipulation using a Quadruped Robot

Teaches a quadruped to push stacks of unsecured objects with its body without toppling them.

Stergios E. Bachoumas et al.

11 papers

Whole-Body Control

MPC, impedance, safety layers, and whole-body coordination.

paper preview PATACON: GPU-Parallel Tangent-Bundle RRT-Connect for Constrained Motion Planning visual

35arXiv

PATACON: GPU-Parallel Tangent-Bundle RRT-Connect for Constrained Motion Planning

Finds fast collision-free G1 whole-body paths by planning many constrained routes on the GPU.

Gahyun Oh et al.

paper preview Adaptive Whole-Body MPC with Optimal Posture Planning for Humanoid Robots under Unknown Payloads visual

36arXiv

Adaptive Whole-Body MPC with Optimal Posture Planning for Humanoid Robots under Unknown Payloads

Keeps a humanoid balanced while lifting unknown loads by adapting posture and force plans online.

Manh Cuong Nguyen et al.

paper preview Task-Space Motion-Force Nonlinear Model Predictive Control for Precise and Compliant Control under Uncertainties visual

37arXiv

Task-Space Motion-Force Nonlinear Model Predictive Control for Precise and Compliant Control under Uncertainties

Balances motion and contact forces so legged robots stay compliant on uncertain terrain.

Quang-Nam Nguyen et al.

paper preview Analytic Robot Self-Model: Learning Screw-Axis Kinematics for Rendering and Model-Based Control visual

38arXiv

Analytic Robot Self-Model: Learning Screw-Axis Kinematics for Rendering and Model-Based Control

Learns an explicit robot body model that can render and control new joint configurations.

Chen Xu et al.

paper preview Modeling and Control of a Rigid-soft Hybrid-link Humanoid Robot visual

39arXiv

Modeling and Control of a Rigid-soft Hybrid-link Humanoid Robot

Models rigid and soft robot links together so a humanoid with prosthetic parts can move compliantly.

Zewen He et al.

paper preview Safe Reinforcement Learning with Receding-Horizon Safety Critic and Differentiable Action Shields visual

40arXiv

Safe Reinforcement Learning with Receding-Horizon Safety Critic and Differentiable Action Shields

Adds a predictive safety shield that corrects unsafe quadruped actions before execution.

Siwei Ju et al.

project clip

41arXiv

Contact-UAN: Unsupervised Actuator Networks as a Residual World Model for Humanoid Control

Learns missing actuator and contact dynamics from a few minutes of walking data to improve humanoid control.

Nolan Fey et al.

paper preview Inertially Significant Humanoid Teleoperation via Whole Body Model Predictive Control and Online Parameter Estimation visual

42arXiv

Inertially Significant Humanoid Teleoperation via Whole Body Model Predictive Control and Online Parameter Estimation

Keeps a humanoid balanced under unknown asymmetric loads by estimating payloads during teleoperation.

James Foster et al.

paper preview LEAP: Gradient Flow Augmented Lagrangian Trajectory Optimization for Contact Constrained Robots visual

43arXiv

LEAP: Gradient Flow Augmented Lagrangian Trajectory Optimization for Contact Constrained Robots

Optimizes contact-rich robot motions while continuously enforcing force and contact constraints.

César E. Ramos Chuquiure et al.

paper preview Planning to Learn: Real-Time Robust Adaptive Control on the GPU with Guaranteed Future Learning visual

44arXiv

Planning to Learn: Real-Time Robust Adaptive Control on the GPU with Guaranteed Future Learning

Plans quadruped actions that control the robot now while improving what it can learn next.

Anutam Srinivasan et al.

paper preview Manifold Generative Annealing for Contact Constraint-aware Impedance Control with Reinforcement Learning Prior visual

45arXiv

Manifold Generative Annealing for Contact Constraint-aware Impedance Control with Reinforcement Learning Prior

Refines RL-proposed humanoid motions to satisfy contact and impedance constraints.

Zhilin He et al.

21 papers

Training Infrastructure

Benchmarks, datasets, rewards, simulators, and tooling.

paper preview Reinforcement Learning via Fixed-point Flow Matching visual

46arXiv

Reinforcement Learning via Fixed-point Flow Matching

Trains locomotion policies by fitting smooth action flows directly from critic feedback.

Denis Blessing et al.

paper preview SHOES: Sim-to-real Hardware-adaptation via On-policy Expansion of Skills for Legged Robots visual

47arXiv

SHOES: Sim-to-real Hardware-adaptation via On-policy Expansion of Skills for Legged Robots

Adapts a frozen quadruped policy to hardware changes, such as wearing shoes, through short on-robot updates.

Kota Tsuruhara et al.

project clip

48arXiv

Test-Time Compute for Humanoid Behavior Foundation Models

Improves a humanoid behavior model at run time by testing candidate commands before executing them.

Weiji Xie et al.

paper preview Chunk-wise Model-based Policy Optimization visual

49arXiv

Chunk-wise Model-based Policy Optimization

Improves locomotion policies by learning dynamics and optimizing actions in short chunks.

Weixi Song et al.

paper preview Reinforcement Learning from Real-Robot Interaction without High-Quality State Estimates visual

50arXiv

Reinforcement Learning from Real-Robot Interaction without High-Quality State Estimates

Trains robot policies from real interactions even when accurate state estimates are unavailable.

Ryan Gao et al.

project clip

51arXiv

Agentic Robot Creation through Generation, Verification, and Revision

Lets an agent design, test, and revise legged robot skills automatically.

Xuebing Li et al.

paper preview Code2Worlds-2: Physically Plausible Agent Harness for Embodied World Simulation visual

52arXiv

Code2Worlds-2: Physically Plausible Agent Harness for Embodied World Simulation

Builds physically testable MuJoCo worlds so agents can train and evaluate robot behaviors reliably.

Haoyu Zhang et al.

paper preview Improving Policy Learning from Delayed Interventions visual

53arXiv

Improving Policy Learning from Delayed Interventions

Learns better humanoid policies from human corrections that arrive late.

Xiru Zhu et al.

paper preview Generative Flow on Symbolic Abstractions for Embodiment Capability Discovery visual

54arXiv

Generative Flow on Symbolic Abstractions for Embodiment Capability Discovery

Discovers progressively harder robot skills by reasoning over what each embodiment can physically do.

Youwei Yu et al.

paper preview Barrier-Constrained Guidance For Real-time Asynchronous VLA Execution visual

55arXiv

Barrier-Constrained Guidance For Real-time Asynchronous VLA Execution

Keeps asynchronous action chunks aligned with safety limits during biped and legged control.

Wenbo Yu et al.

paper preview CFG-Policy: Unified Diffusion Policy Learning Framework via Classifier-free Guidance visual

56arXiv

CFG-Policy: Unified Diffusion Policy Learning Framework via Classifier-free Guidance

Trains diffusion-based locomotion policies while keeping exploration separate from execution.

Shutong Ding et al.

paper preview Model-Based Online Decision Making via Generative Trajectory Planning visual

57arXiv

Model-Based Online Decision Making via Generative Trajectory Planning

Plans complete action trajectories online to improve simulated locomotion over short horizons.

Haldun Balim et al.

paper preview DR-APO: Decoupled-Representation Advantage-Guided Policy Optimization from Continuous Control to Embodied Navigation visual

58arXiv

DR-APO: Decoupled-Representation Advantage-Guided Policy Optimization from Continuous Control to Embodied Navigation

Improves locomotion policies using one update rule across several policy designs.

Runhua Zhang et al.

paper preview DEFlow: Deep Equilibrium Flow Maximum-Entropy Reinforcement Learning visual

59arXiv

DEFlow: Deep Equilibrium Flow Maximum-Entropy Reinforcement Learning

Trains flow-based locomotion policies with maximum-entropy reinforcement learning.

Ruslan Ageev et al.

project clip

60arXiv

Modeling and Understanding Neural Weights of Humanoid Policies with Diffusion

Generates new humanoid locomotion policies by learning the structure of existing policy weights.

Huixuan Cheng et al.

paper preview Efficient Test-Time Adaptation For Robot Policies visual

61arXiv

Efficient Test-Time Adaptation For Robot Policies

Adapts a locomotion policy during deployment using its critic, without requiring reward signals.

Motasem Alfarra et al.

paper preview λ_0: Scaling Open-World Egocentric Human Data for Humanoid Loco-Manipulation visual

62arXiv

λ_0: Scaling Open-World Egocentric Human Data for Humanoid Loco-Manipulation

Turns large-scale human egocentric data into aligned training data for G1 loco-manipulation.

Lambda-humanoid Team

project visual RP1: open humanoid engineering resources and October 15 CAD source release visual

63GitHub

RP1: open humanoid engineering resources and October 15 CAD source release

Releases editable hardware, manufacturing, and control resources for the RP1 humanoid platform.

RoboParty et al.

project visual Open Sprite Runtime: native actor and software-IMU supervised walking candidate visual

64GitHub

Open Sprite Runtime: native actor and software-IMU supervised walking candidate

Runs supervised humanoid walking with native inference, software IMU sensing, and guarded motor control.

BeijingDynamics et al.

paper preview KODA: Kinematics--Observation Data Alignment for Humanoid Learning from Human Egocentric Demonstration visual

65arXiv

KODA: Kinematics--Observation Data Alignment for Humanoid Learning from Human Egocentric Demonstration

Aligns egocentric human demonstrations with G1 motion, vision, and controller-specific action labels.

Zibo Zhou et al.

paper preview CQDC-Warp: GPU-Accelerated Differentiable Quasi-Dynamic Simulation for Contact-Rich Manipulation visual

66arXiv

CQDC-Warp: GPU-Accelerated Differentiable Quasi-Dynamic Simulation for Contact-Rich Manipulation

Accelerates differentiable contact simulation for recovery, mobile manipulation, and humanoid postures.

Yanchen Ding et al.

8 papers

State Estimation / Perception

Terrain, state, slip, sensors, and local context.

paper preview From Pixels to Executable Physics: Behavior-Grounded World Models for Embodied Action Consequence Prediction visual

67arXiv

From Pixels to Executable Physics: Behavior-Grounded World Models for Embodied Action Consequence Prediction

Predicts how terrain and candidate actions will affect a quadruped before it moves.

Ruo Chen et al.

paper preview LiDAR-based Humanoid Navigation in Dynamic, Non-Convex Environments via Spatio-Temporal Safe Corridors visual

68arXiv

LiDAR-based Humanoid Navigation in Dynamic, Non-Convex Environments via Spatio-Temporal Safe Corridors

Guides Digit through moving obstacles by turning LiDAR into safe corridors and footstep plans.

Shubham S. Kumbhar et al.

paper preview PointWalker: Mapless Quadruped Locomotion Directly From Raw LiDAR Point Clouds visual

69arXiv

PointWalker: Mapless Quadruped Locomotion Directly From Raw LiDAR Point Clouds

Lets a quadruped navigate rough spaces directly from raw LiDAR points without building a map.

Sicen Li et al.

project clip

70arXiv

Beyond Fixed Visual History: Learning Motion-Adaptive Temporal Receptive Fields for Complex-Terrain Quadruped Locomotion

Lets a quadruped adjust how much visual history it uses as terrain and motion change.

Hao Zhang et al.

paper preview GLST: Global-Local Spatio-Temporal Memory Decomposition Framework for Online Chained-Goal Navigation visual

71arXiv

GLST: Global-Local Spatio-Temporal Memory Decomposition Framework for Online Chained-Goal Navigation

Helps quadrupeds navigate chained goals by separating route memory from immediate terrain memory.

Yousef Liang et al.

paper preview Learning Reachability-Based Safety Representations for Real-Time Robot Navigation in Unknown and Dynamic Environments visual

72arXiv

Learning Reachability-Based Safety Representations for Real-Time Robot Navigation in Unknown and Dynamic Environments

Learns balance-safe regions that help quadrupeds navigate moving environments without tipping.

Bojan Derajić

paper preview Sensing via whole-body interactions with complex terrains: physical insights from robotic models visual

73arXiv

Sensing via whole-body interactions with complex terrains: physical insights from robotic models

Explains how whole-body contact can help legged robots sense and cross complex terrain.

Chen Li

paper preview E2DRL: End-to-End Differentiable Reinforcement Learning for 3D Obstacle Avoidance in Quadruped Robots visual

74arXiv

E2DRL: End-to-End Differentiable Reinforcement Learning for 3D Obstacle Avoidance in Quadruped Robots

Teaches a quadruped to crawl under and steer around obstacles using depth input.

Cheng Fan et al.