Paper Sweeps

Humanoid Papers from June 14 to June 28, 2026

7Locomotion12Motion Tracking11Loco-Manipulation7Whole-Body Control8Training Infrastructure5State Estimation / Perception1Other

7 papers

Locomotion

Terrain, gait, robustness, and agile movement.

paper preview ADAPT: Analytical Disturbance-Aware Policy Training for Humanoid Locomotion visual

01arXiv

ADAPT: Analytical Disturbance-Aware Policy Training for Humanoid Locomotion

Keeps humanoids walking through pushes, payloads, and surprise contact by estimating the disturbance and correcting the gait.

Bofan Lyu et al.

paper preview Steering Generative Reinforcement Learning into Stable Robotic Controller visual

02arXiv

Steering Generative Reinforcement Learning into Stable Robotic Controller

Turns generative robot policies into steadier walking controllers by steering training toward stable actions.

Yixuan Wang et al.

paper preview Elastic ODYN: Differentiable Optimization for Infeasible Control and Learning in Robotics visual

03arXiv

Elastic ODYN: Differentiable Optimization for Infeasible Control and Learning in Robotics

Keeps robot planners useful when contact, constraints, or objectives make the normal control math break down.

Aristotelis Papatheodorou et al.

project clip

04arXiv

Robust Fall Recovery for Armless Bipedal-Wheeled Robots Via Force-Guided Learning

Teaches an armless bipedal-wheeled robot to get back up by applying useful force to the ground.

Haidong Hou et al.

project clip

05arXiv

SWAP: Symmetric Equivariant World-Model for Agile Robot Parkour

Teaches quadruped parkour moves like gap jumping and box climbing with a symmetry-aware prediction model.

Kaixin Lan et al.

project clip

06arXiv

CTS-MoE: Implicit Terrain Adaptation via Mixture-of-Experts for Perceptive Locomotion

Adapts a legged robot's walking style across stairs, gaps, slopes, and obstacles using depth perception.

Francisco Affonso et al.

project visual SCAN-Planner: Spatial Collision-Aware Local Planning for Route-Guided Long-Range Quadruped Navigation visual

07arXiv

SCAN-Planner: Spatial Collision-Aware Local Planning for Route-Guided Long-Range Quadruped Navigation

Plans quadruped routes through clutter by checking the robot's whole body for collisions.

Han Zheng et al.

12 papers

Motion Tracking

Reference motion, imitation, retargeting, and behavior priors.

paper preview VENOM: Versatile Embodied Network for Omni-bodied Motion Tracking visual

08arXiv

VENOM: Versatile Embodied Network for Omni-bodied Motion Tracking

Copies full-body motions across different humanoid bodies with one shared motion-tracking system.

Siddharth Padmanabhan et al.

paper preview Video-Based Optimal Transport for Feedback-Efficient Offline Preference-Based Reinforcement Learning visual

09arXiv

Video-Based Optimal Transport for Feedback-Efficient Offline Preference-Based Reinforcement Learning

Uses videos to teach robots which motions people prefer, with fewer hand-labeled examples.

Tung M. Luu et al.

paper preview Kine2Go: Kinematic dataset for the Unitree Go2 robot with diverse gaits and motions visual

10arXiv

Kine2Go: Kinematic dataset for the Unitree Go2 robot with diverse gaits and motions

Provides Go2 motion examples so legged robots can train from many walking and gait patterns.

Wladyslaw Palucki et al.

project clip

11arXiv

Generalizing from References using a Multi-Task Reference and Goal-Driven RL Framework

Uses reference motions during training so a G1 humanoid can perform goal-driven parkour skills later.

Jiashun Wang et al.

paper preview Comparative Study on Agility, Efficiency, and Impact Absorption of Bipedal Robots with Active Toes visual

12arXiv

Comparative Study on Agility, Efficiency, and Impact Absorption of Bipedal Robots with Active Toes

Studies whether active toes make biped robots more agile, efficient, and impact-resistant.

Joong-Gil Kim et al.

paper preview Mind Your Steps: A General Learning Framework for Accurate Humanoid Foothold Tracking visual

13arXiv

Mind Your Steps: A General Learning Framework for Accurate Humanoid Foothold Tracking

Places humanoid feet accurately on target footholds while higher-level planners choose the path.

Alessandro Montenegro et al.

paper preview Learning Asynchronous Upper-body Task-space Trajectory Tracking Policy for Humanoid Robots visual

14arXiv

Learning Asynchronous Upper-body Task-space Trajectory Tracking Policy for Humanoid Robots

Lets a humanoid follow sparse upper-body task commands while filling in the balance motion.

Yumeng Liu et al.

paper preview MAPL: Multi-Objective Preference Learning for Robot Locomotion visual

15arXiv

MAPL: Multi-Objective Preference Learning for Robot Locomotion

Learns walking rewards from preferences so robots need less hand-tuned reward design.

Xiyue Chen et al.

paper preview BFMTrack: Latent Sequence Optimization for Physics-Based Motion Tracking with Behavioral Foundation Models visual

16arXiv

BFMTrack: Latent Sequence Optimization for Physics-Based Motion Tracking with Behavioral Foundation Models

Tracks full or partial humanoid motion references while keeping the movement physically plausible.

Thomas Rupf et al.

project visual HumanoidUMI: Bridging Robot-Free Demonstrations and Humanoid Whole-Body Manipulation visual

17arXiv

HumanoidUMI: Bridging Robot-Free Demonstrations and Humanoid Whole-Body Manipulation

Collects human whole-body demonstrations without the robot present and retargets them into humanoid skills.

Hongwu Wang et al.

project visual PressMimic: Pressure-Guided Motion Capture and Control for Humanoid Robot Imitation visual

18arXiv

PressMimic: Pressure-Guided Motion Capture and Control for Humanoid Robot Imitation

Uses foot pressure so humanoids imitate both body poses and real contact with the floor.

Yi Lu et al.

project visual Perceptive Behavior Foundation Model: Adapting Human Motion Priors to Robot-Centric Terrain visual

19arXiv

Perceptive Behavior Foundation Model: Adapting Human Motion Priors to Robot-Centric Terrain

Adapts human-motion skills to robot-centered terrain so humanoids adjust foot placement while moving expressively.

Zifan Wang et al.

11 papers

Loco-Manipulation

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

paper preview Mobile Pedipulation for Object Sliding via Hierarchical Control on a Wheeled Bipedal Robot visual

20arXiv

Mobile Pedipulation for Object Sliding via Hierarchical Control on a Wheeled Bipedal Robot

Lets a wheeled biped push and slide objects with its legs while staying balanced.

Yue Qin et al.

paper preview HALOMI: Learning Humanoid Loco-Manipulation with Active Perception from Human Demonstrations visual

21arXiv

HALOMI: Learning Humanoid Loco-Manipulation with Active Perception from Human Demonstrations

Teaches a humanoid to move, look, and use its hands by adapting human demonstration videos.

Zehui Zhao et al.

paper preview DynaRetarget: Dynamically-Feasible Retargeting using Sampling-Based Trajectory Optimization visual

22arXiv

DynaRetarget: Dynamically-Feasible Retargeting using Sampling-Based Trajectory Optimization

Turns human object-interaction motions into physically usable trajectories for a G1 humanoid.

Victor Dhedin et al.

project clip

23arXiv

CoorDex: Coordinating Body and Hand Priors for Continuous Dexterous Humanoid Loco-Manipulation

Combines walking and dexterous hand skills so a humanoid can move and manipulate continuously.

Sikai Li et al.

project visual FT-WBC: Learning Fault-Tolerant Whole-Body Control for Legged Loco-Manipulation visual

24arXiv

FT-WBC: Learning Fault-Tolerant Whole-Body Control for Legged Loco-Manipulation

Keeps a legged robot with an arm moving and manipulating even when some motors fail.

Yudong Zhong et al.

project visual OpenHLM: An Empirical Recipe for Whole-Body Humanoid Loco-Manipulation visual

25arXiv

OpenHLM: An Empirical Recipe for Whole-Body Humanoid Loco-Manipulation

Gives a practical recipe for teaching humanoids whole-body movement and manipulation skills.

Yingdong Hu et al.

paper preview WOLF-VLA: Whole-Body Humanoid Optimal Locomotion Framework for Vision-Language-Action Learning visual

26arXiv

WOLF-VLA: Whole-Body Humanoid Optimal Locomotion Framework for Vision-Language-Action Learning

Builds vision, body-sensor, and language data for humanoids that move with task instructions.

Melya Boukheddimi et al.

paper preview Humanoid-DART: Humanoid Loco-Manipulation using Diffusion-guided Augmentation through Relabeling and Tracking visual

27arXiv

Humanoid-DART: Humanoid Loco-Manipulation using Diffusion-guided Augmentation through Relabeling and Tracking

Expands a small set of humanoid manipulation demos into more goal-conditioned training tasks.

Pranav Debbad et al.

paper preview A System for Fast, Resilient, and Adaptable Loco-Manipulation Behaviors on Humanoid Robots visual

28arXiv

A System for Fast, Resilient, and Adaptable Loco-Manipulation Behaviors on Humanoid Robots

Lets people author practical humanoid behaviors around objects, such as moving through doors.

Duncan William Calvert

project visual OmniContact: Chaining Meta-Skills via Contact Flow for Generalizable Humanoid Loco-Manipulation visual

29arXiv

OmniContact: Chaining Meta-Skills via Contact Flow for Generalizable Humanoid Loco-Manipulation

Chains humanoid loco-manipulation skills by planning when feet, hands, or body parts should touch the world.

Runyi Yu et al.

project visual TaskNPoint: How to Teach Your Humanoid to Hit a Backhand in Minutes visual

30arXiv

TaskNPoint: How to Teach Your Humanoid to Hit a Backhand in Minutes

Teaches a G1 humanoid a dynamic skill like a tennis backhand from a single video.

Blake Werner et al.

7 papers

Whole-Body Control

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

paper preview Whole-Body Impedance Model Predictive Control for Safe Physical Human-Robot Interaction on Floating-Base Platforms visual

31arXiv

Whole-Body Impedance Model Predictive Control for Safe Physical Human-Robot Interaction on Floating-Base Platforms

Balances a humanoid during physical interaction by coordinating whole-body motion and changing contact forces.

Yongyan Cao

paper preview ZiMPedance: Impedance-Aware ZMP Modeling and Control for Payload Carrying with Quadruped Robots visual

32arXiv

ZiMPedance: Impedance-Aware ZMP Modeling and Control for Payload Carrying with Quadruped Robots

Helps quadrupeds carry payloads more safely by accounting for how the load moves with the robot.

Giovanni B. Dessy et al.

paper preview HumanoidArena: Benchmarking Egocentric Hierarchical Whole-body Learning visual

33arXiv

HumanoidArena: Benchmarking Egocentric Hierarchical Whole-body Learning

Tests whether humanoid task decisions can turn into stable whole-body motion in simulation.

Taowen Wang et al.

paper preview TurboMPC: Fast, Scalable, and Differentiable Model Predictive Control on the GPU visual

34arXiv

TurboMPC: Fast, Scalable, and Differentiable Model Predictive Control on the GPU

Makes fast robot planning run on GPUs so controllers can make quicker movement decisions.

Gabriel Bravo-Palacios et al.

paper preview Mixture-of-Experts RL for Fault-Tolerant Legged Locomotion visual

35arXiv

Mixture-of-Experts RL for Fault-Tolerant Legged Locomotion

Keeps a quadruped walking after actuator damage by routing control to the right expert behavior.

Giulio Turrisi et al.

paper preview RGB: RL Guided Whole-Body MPPI for Humanoid Control visual

36arXiv

RGB: RL Guided Whole-Body MPPI for Humanoid Control

Guides online humanoid whole-body control with an existing walking controller instead of retraining.

Yunsoo Seo et al.

paper preview MPC-Injection: Biasing Off-Policy Locomotion RL Toward Controller-Induced Behavior Basins visual

37arXiv

MPC-Injection: Biasing Off-Policy Locomotion RL Toward Controller-Induced Behavior Basins

Steers robot learning toward walking patterns that a model-based controller already knows can work.

Roy Xing et al.

8 papers

Training Infrastructure

Benchmarks, datasets, rewards, simulators, and tooling.

paper preview SRL: Combining SLIP Model and Reinforcement Learning for Agile Robotic Jumping visual

38arXiv

SRL: Combining SLIP Model and Reinforcement Learning for Agile Robotic Jumping

Teaches legged robots to jump over obstacles and stairs by combining simple physics with practice.

Xiaowen Hu et al.

paper preview ForEnt: A Multi-Modal Dataset for Characterizing Quadruped Robot Entrapments in Forest Environments visual

39arXiv

ForEnt: A Multi-Modal Dataset for Characterizing Quadruped Robot Entrapments in Forest Environments

Records forest entrapment failures so quadrupeds can learn to detect and avoid getting stuck.

Natapat Kirdwichai et al.

project visual LoComposition: Terrain-Adaptive Energy-Efficient Quadruped Locomotion without Gait Priors visual

40arXiv

LoComposition: Terrain-Adaptive Energy-Efficient Quadruped Locomotion without Gait Priors

Trains a quadruped to cross rough terrain without hand-designed gait preferences.

Loukas Kordos et al.

paper preview Simulating Robotic Locomotion in Sand: Resistive Force Theory in an Open-Source Physics Engine visual

41arXiv

Simulating Robotic Locomotion in Sand: Resistive Force Theory in an Open-Source Physics Engine

Adds sand simulation so legged robots can test how their feet push through soft ground.

Ryan Walker Brown et al.

project clip

42arXiv

DF-ExpEnse: Diffusion Filtered Exploration for Sample Efficient Finetuning

Improves a pretrained robot controller with fewer trials by filtering for more useful exploration.

Calvin Luo et al.

paper preview CRAX: Fast Safe Reinforcement Learning Benchmarking visual

43arXiv

CRAX: Fast Safe Reinforcement Learning Benchmarking

Benchmarks safe robot training by testing when simulated walking agents break safety constraints.

Tristan Tomilin et al.

paper preview DynaWM: Dynamics-Aware Distillation with World Model and Momentum Targets for Smooth Locomotion over Continuous Stairs visual

44arXiv

DynaWM: Dynamics-Aware Distillation with World Model and Momentum Targets for Smooth Locomotion over Continuous Stairs

Helps a bipedal-wheeled robot climb long staircases smoothly with less sensing at runtime.

Haidong Hou et al.

project visual StairMaster: Learning to Conquer Risky Hollow Stairs for Agile Quadrupedal Robots visual

45arXiv

StairMaster: Learning to Conquer Risky Hollow Stairs for Agile Quadrupedal Robots

Trains a quadruped to climb hollow stairs while avoiding leg traps and bad foot placements.

Xincheng Tang et al.

5 papers

State Estimation / Perception

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

paper preview GAIT: Legged Robot Proprioceptive State Estimation with Attention over Inertial-Leg Tokens visual

46arXiv

GAIT: Legged Robot Proprioceptive State Estimation with Attention over Inertial-Leg Tokens

Estimates a legged robot's body motion by learning which onboard body and leg sensors to trust.

Young-Rang Seo et al.

paper preview Proprioceptive Invariant State Estimation for Humanoid Robots on Non-Inertial Ground visual

47arXiv

Proprioceptive Invariant State Estimation for Humanoid Robots on Non-Inertial Ground

Estimates humanoid body motion on moving ground using only torso, foot, and joint sensors.

Falak Mandali et al.

paper preview SlipSense: Multimodal Sensing for Online Slip Detection in Legged Robots visual

48arXiv

SlipSense: Multimodal Sensing for Online Slip Detection in Legged Robots

Detects foot slip early with a sensorized foot so a legged robot can react before falling.

Iris Szu-Yao Liu et al.

paper preview LP-NavOA: Integrated Local Navigation and Obstacle Avoidance for Humanoid Robots under Limited Perception visual

49arXiv

LP-NavOA: Integrated Local Navigation and Obstacle Avoidance for Humanoid Robots under Limited Perception

Helps a humanoid avoid obstacles with limited short-range sensing while still moving toward the goal.

Yukun Luo et al.

paper preview Learning Perceptive Platform Adaptive Locomotion Controllers for Quadrupedal Robots visual

50arXiv

Learning Perceptive Platform Adaptive Locomotion Controllers for Quadrupedal Robots

Trains walking controllers that adapt across different quadruped body shapes and perception setups.

David Rytz et al.

1 paper

Other

Adjacent humanoid robotics methods.

paper preview TEXEDO: Test Time Scaling for Controller-aware Language-conditioned Humanoid Motion Generation visual

51arXiv

TEXEDO: Test Time Scaling for Controller-aware Language-conditioned Humanoid Motion Generation

Turns text prompts into robot-feasible humanoid motions by rejecting moves the controller cannot follow.

Jianuo Cao et al.