project clip
01arXiv
SkateNav: Learning Energy-Efficient Navigation through Adaptive Skating on Passive-Wheel Quadrupedal Robots
Teaches wheel-legged quadrupeds when to push and coast for efficient navigation.
Paper Sweeps
Humanoid Papers from October 18 to October 24, 2026
14Locomotion13Motion Tracking16Loco-Manipulation6Whole-Body Control14Training Infrastructure2State Estimation / Perception
14 papers
Locomotion
Terrain, gait, robustness, and agile movement.
paper preview
02arXiv
Validation of Human Motor Development Scales with Humanoid Robots
Teaches humanoids to stand up by progressing through support poses instead of copying demonstrations.
paper preview
03arXiv
SWAP: Test-Time Steerable Humanoid Whole-Body Collision Avoidance
Lets a humanoid choose safe whole-body motions around obstacles based on the current goal.
paper preview
04arXiv
Anticipatory Gait Switching for Quadrupeds via Look-Ahead MDP Planning
Switches quadruped gaits before terrain changes arrive so the robot stays stable and efficient.
project clip
05arXiv
FCTS: Failure-Guided Curriculum Tree Search Over Generated Terrain for Legged Parkour
Builds harder parkour terrain around each failure so a legged robot practices what it currently cannot do.
project clip
06arXiv
Adaptive Traversal Tubes for Posture-Aware Navigation of Bipedal Humanoids
Changes a biped's posture to fit narrow passages while keeping the existing walking controller.
paper preview
07arXiv
Revealing and Bridging the Physical Executability Gap in Vision-Language Navigation
Adjusts vision-language navigation paths so a quadruped chooses routes it can physically traverse.
paper preview
08arXiv
PASS-VLN: Passability-Aware Local-Command Vision-and-Language Navigation
Guides a quadruped through tight passages using body-width-aware depth maps and local language commands.
project visual
09arXiv
Watch Your Step: Anticipated Perception for Humanoid Locomotion
Predicts future foot placements and focuses terrain sensing only where a humanoid is about to step.
paper preview
10arXiv
Do Earth-Trained Humanoid Controllers Walk on the Moon? Gravity, Terrain, and Soil Transfer
Trains humanoid walking that transfers between Earth and lunar gravity, terrain, and soil.
paper preview
11arXiv
End-to-End Reinforcement Learning for Quadrupedal Jumping in Lunar Gravity
Teaches a quadruped to launch, control its body in flight, land, and recover in lunar gravity simulation.
project visual
12arXiv
Humanoid Mobility via Attention-based Contacts
Teaches humanoids to use hands, feet, knees, and other contacts to move across difficult terrain.
paper preview
13arXiv
MPBRS: Multi-Potential-Based Reward Shaping for Unified Reward Configuration in Humanoid Locomotion
Reuses one reward-shaping recipe across many simulated humanoid bodies and walking goals.
paper preview
14arXiv
Sensitive dependence of the motion of a legged robot on granular media
Shows how gait timing and sand packing change a legged robot from rotary walking to slow swimming.
13 papers
Motion Tracking
Reference motion, imitation, retargeting, and behavior priors.
paper preview
15arXiv
RoboMMT: Learning Stable Tracking of Modulated Expressive and Interactive Robot Motion
Keeps expressive humanoid gestures stable and lets them react to nearby people.
project clip
16arXiv
Vision as Proprioception: Estimating Humanoid Robot Pose from Video
Turns ordinary humanoid video into estimated joint motion for control and data generation.
project clip
17arXiv
Weakly Supervised Discovery of Motion Styles for Compositional Robot Control
Discovers selectable walking styles from motion recordings with only weak labels.
project clip
18arXiv
BeJEPA: A Goal-Conditioned Humanoid Behavior World Model via Unsupervised RL
Learns a humanoid behavior model that predicts useful actions and can refine them before execution.
project clip
19arXiv
Fast Motion Retargeting Pipeline for Scalable Humanoid Robot Learning
Retargets human motion quickly, then spends dynamics checks only on motions that need them.
project clip
20arXiv
SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model
Learns a humanoid world model and tracking policy together for fast whole-body motion control.
project clip
21arXiv
Joints, Camera, Action! Co-Generating 3D Motion and Video for Arbitrary Rigs
Generates matching video and robot-rig motion so arbitrary bodies, including G1, can follow the same scene.
project clip
22arXiv
HuMiT: Low-Latency Whole-Body Humanoid Teleoperation via Compact Reference Tracking
Drives low-latency whole-body humanoid teleoperation from a compact current-frame reference.
paper preview
23arXiv
SMR: Learning streaming motion retargeting for humanoids from human semantic
Learns real-time humanoid retargeting from motions that were successfully executed on the robot.
paper preview
24arXiv
One Policy, Many Styles: Where to Aggregate Preferences for Steerable Humanoid Control?
Makes one humanoid policy respond to different style preferences by choosing where to combine objectives.
project visual
25arXiv
Robot Motion Transition with Human Motion Prior
Creates smooth, executable transitions between separate human motion clips for humanoid tracking.
paper preview
26arXiv
Motion Retargeting with Spatiotemporal Preferences for Humanoid Loco-Manipulation
Retargets human motion by focusing on the frames and body parts that matter most for object interaction.
paper preview
27arXiv
PhysibleMDM: Enforcing Inference-Time Physical Feasibility Constraints on Motion Diffusion Models via Noise Optimization
Makes generated biped motions more physically feasible at inference time without retraining the motion model.
16 papers
Loco-Manipulation
Whole-body task contact: feet, torso, hands, and objects.
paper preview
28arXiv
HAF-VLA: Hierarchical Action Flow Framework for Humanoid Vision-Language-Action Control
Plans the humanoid's base, posture, and arms in stages so it can follow language commands.
paper preview
29arXiv
OmniInteract: Towards Unified Whole-Body Interaction Control for Dual Humanoids
Coordinates two humanoids through shared contact so they can interact without losing balance.
project clip
30arXiv
Humanoid SRL: Augmenting Humanoids with Supernumerary Limbs for Coordinated Loco-Manipulation
Gives humanoids an extra arm for wider reach, carrying support, and coordinated movement.
paper preview
31arXiv
DiPLO: Self-distilled Diffusion Planning for Loco-manipulation
Plans humanoid movement and manipulation with diffusion, then distills it into a faster controller.
project clip
32arXiv
Action-Aligned Latent Planning for Versatile Humanoid Loco-Manipulation
Plans humanoid movement through compact action-aware predictions instead of replaying fixed motions.
project clip
33arXiv
Solving a Giant Rubik's Cube with a Quadruped Robot
Teaches a quadruped to brace and turn a giant Rubik's Cube using both front legs and whole-body balance.
paper preview
34arXiv
Language-Forcing: Advancing Humanoid Loco-Manipulation via Semantic Alignment of Whole-Body World Action Models
Aligns a humanoid world model with language so its movement reflects task progress, not just visual change.
paper preview
35arXiv
DUET: Decoupled Upper- and Lower-Body Execution for Task-Agnostic Humanoid Loco-Manipulation
Lets one arm policy handle tasks while a separate lower-body controller keeps the humanoid balanced.
paper preview
36arXiv
Learning Whole-Body Kinetic Coordination for Wheeled-Quadruped Long-Range Throwing
Teaches a wheeled quadruped to pitch and coordinate its body for longer throws on varied terrain.
paper preview
37arXiv
Reinforcement Learning Based Multi-Terrain Adaptive Whole-Body Control for a Quadruped Manipulator
Coordinates a quadruped and arm with one controller while moving and manipulating over uneven terrain.
paper preview
38arXiv
CoEvolve: Self-Evolving Code Policies from Skills to Strategies for Humanoid Robots
Lets coding agents improve cooperative humanoid tasks by testing and revising executable robot programs.
paper preview
39arXiv
Scaling Online Exploration for Robot Agents
Lets a coding agent learn humanoid tasks online, then save successful perception and control steps as reusable programs.
paper preview
40arXiv
LocoPilot: Single-Operator Bilateral Teleoperation for Loco-Manipulation Policy Learning
Collects loco-manipulation demonstrations with one operator using the same actions the quadruped executes at deployment.
project clip
41arXiv
VLM-Guided Contact Retargeting and Generative Partner Modeling for Humanoid-Human Interaction
Preserves touch points and predicts a human partner's hidden motion so a humanoid can interact while walking.
project visual
42arXiv
SP-HLM: Sparse-Pose-Driven Motion Generation and Whole-Body Control for Humanoid Loco-Manipulation
Generates whole-body humanoid motion from a few task poses while preserving task geometry.
project clip
43arXiv
Memory before Goal: Role of Pre-History Memory for Humanoid Loco-Manipulation
Gives a humanoid memory of earlier object events so it can finish whole-body tasks after cues disappear.
6 papers
Whole-Body Control
MPC, impedance, safety layers, and whole-body coordination.
paper preview
44arXiv
Bio-inspired Thrust-Assisted Trotting on a legged-Aerial Robot via Real-Time QP-MPC
Uses small thrusters to help a biped trot while a fast planner balances foot forces.
paper preview
45arXiv
Learning Structured Locomotion References for Optimization-Based Whole-Body Control
Learns useful walking references while keeping an optimization-based whole-body controller in charge.
paper preview
46arXiv
Adaptive Rejection of Unknown Hybrid Sinusoidal Disturbances with Application to Humanoid Walking under Unknown Ground Motion
Keeps a humanoid walking on moving ground by adapting to unknown repeated floor motion.
paper preview
47arXiv
Design and Control of a Multimodal Legged-Aerial Robot Through Separate Control Contraction Metrics
Coordinates legs and thrusters with separate stability measures for a robot that can walk and fly.
paper preview
48arXiv
Temporal Difference Learning of Probabilistic Reach-Avoid Value Functions
Predicts fall risk and switches to safer actions before a legged robot enters a dangerous state.
paper preview
49arXiv
Test-Time Motion Steering for Perceptive Humanoid Fall Recovery
Steers a humanoid's recovery motion at run time so it can use vision to get up from a fall.
14 papers
Training Infrastructure
Benchmarks, datasets, rewards, simulators, and tooling.
paper preview
50arXiv
CusADi2: Latency-Optimized Batch Computation of Robot Kinematics, Jacobians, and Hessians on the GPU
Speeds up robot kinematics and derivatives on GPUs for faster planning and control.
paper preview
51arXiv
Towards Reinforcement Learning for Space Robotics: Fast Training and Reliable Real-World Transfer
Trains quadruped space-robot skills in minutes and transfers selected behaviors to real hardware.
project clip
52arXiv
Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience
Shares predictive training across human and robot video so different bodies can learn loco-manipulation.
paper preview
53arXiv
Vector-Quantized World Model: Latent Feedback Motion Planning for Robot Manipulation
Plans manipulation through discrete visual states while using feedback to stay on course.
project clip
54arXiv
Has GPT-6-Astra Achieved Embodied Intelligence?
Tests whether a general model can issue useful humanoid body and walking commands under fixed controllers.
project clip
55Project
FloMotion: From prompt to policy
Turns prompts or video into retargeted robot motion, rewards, and trained legged policies in a browser workflow.
project clip
56GitHub
Isaac Lab 3.0: General-Availability Robot-Learning Framework Release
Provides one framework for training, testing, and exporting robot policies across simulation backends.
paper preview
57arXiv
Steerable Pluralistic Policy Alignment from Anonymized Crowd Feedback
Learns personalized simulated humanoid styles from anonymous preferences about direction, arm motion, and energy use.
paper preview
58arXiv
Knowing When to Stop Thinking: A Dissection of Expected Free Energy as a Test-Time Compute Controller for Latent World-Model Planning
Stops world-model planning once more thinking is unlikely to improve the robot's action.
project clip
59arXiv
Asymmetric On-Policy Distillation Learns In-Context Exploration
Teaches exploration from privileged training information while keeping the deployed legged policy independent.
paper preview
60arXiv
Make it Walk: Computational Design of LEGO Walkers
Designs LEGO leg shapes and servo timing that produce customizable four- and six-legged walkers.
project clip
61arXiv
TARSphere: A Robot-Oriented Full-Stack Framework for Long-Horizon Embodied Agents in Evolving Physical and Social Environments
Tests long-horizon G1 service agents across walking, manipulation, falls, collisions, and changing human requests.
project clip
62arXiv
LocoManiLab: A Unified Testbed for Data, Policies, Controllers, and Deployment in Humanoid Loco-Manipulation
Compares data, motion generators, policies, controllers, and deployment choices for humanoid household tasks.
paper preview
63arXiv
Learning Robot Behavior Priors from Non-Robotic Synthetic Data
Learns useful quadruped action patterns from synthetic non-robot data for offline command prediction.
2 papers
State Estimation / Perception
Terrain, state, slip, sensors, and local context.
project visual
64arXiv
Learning Wheel Motion Uncertainty for Degeneracy-Aware LiDAR-Inertial Odometry on Quadrupedal Wheel-Legged Robots
Learns when wheel motion is unreliable so a wheel-legged quadruped can estimate its path in poor LiDAR geometry.
paper preview
65arXiv
A2I-Calib++: A Kinematics-Driven Anti-noise Active Calibration Framework for Whole-Body Multi-IMU Systems on Legged Robots
Calibrates several body-mounted IMUs by choosing useful motions and accounting for rolling contact.