paper preview
01arXiv
Tac4Loco: Learning Spatiotemporal Plantar Pressure Representations for Humanoid Locomotion
Uses pressure under a G1 humanoid's feet so it can balance and track commands on slopes, gravel, foam, and narrow supports.
Paper Sweeps
Humanoid Papers from August 16 to August 22, 2026
6Locomotion4Motion Tracking5Loco-Manipulation3Whole-Body Control3Training Infrastructure
6 papers
Locomotion
Terrain, gait, robustness, and agile movement.
paper preview
02arXiv
Trajectory-Level Automatic Curriculum Learning for Legged Locomotion on Unstructured Terrain
Builds terrain lessons for a Go2 automatically so the robot practices jumps and rough paths at the right difficulty.
project visual
03arXiv
Robot-Body-Aware Traversal Risk Graph Planning for Wheeled-Legged Robots in Complex Terrain
Plans safer routes for a wheeled-legged robot by checking how its full body fits through rough terrain.
project clip
04arXiv
Robust Brachiation on a Life-Sized Dual-Arm Robot Using Waypoint-Guided Reinforcement Learning
Teaches a life-sized dual-arm robot to swing between bars using waypoint-guided practice and recovery retries.
paper preview
05arXiv
Learning Highly Dynamic Skills Transition for Quadruped Jumping Through Constrained Space
Teaches a quadruped to pick and switch jumping skills so it can pass through narrow openings.
paper preview
06arXiv
MILD: Tractable Terrain Modeling for Learning Improved Bipedal Locomotion on Deformable Surfaces
Models deformable ground so a physical biped can adjust its walking across soft and stiff surfaces.
4 papers
Motion Tracking
Reference motion, imitation, retargeting, and behavior priors.
project visual
07arXiv
HumanTracker: Towards Comprehensive and Human-Aligned Motion Tracking Benchmark
Measures how well humanoids track motions with a large benchmark and a score that better matches human judgment.
paper preview
08arXiv
HiPHI: A Large-Scale Benchmark for High-Precision Human Motion and Object-Interaction
Collects high-precision human motion and object-interaction data so G1 tracking policies can scale to richer skills.
project visual
09arXiv
GigaBrain-WBC-0.5: A Behavior World Model for Robust Whole-Body Control with Environment Interaction
Uses a behavior world model so humanoids can track motions, interact with terrain, and recover from falls.
project visual
10arXiv
Towards Professional Tennis Styles for Humanoid Robots with Adaptive Motion Planning and Tracking
Turns broadcast tennis motion into humanoid rally and serve skills with adaptive planning and tracking.
5 papers
Loco-Manipulation
Whole-body task contact: feet, torso, hands, and objects.
paper preview
11arXiv
HAF: Adapting Generalist VLAs to Humanoid Whole-Body Loco-manipulation via Hierarchical Action Flow and Spectral Latent RL
Adapts a general robot VLA into a humanoid controller that coordinates walking, posture, and manipulation actions.
project clip
12arXiv
FetchMan: Learning Visual Humanoid Loco-Manipulation Policies from Simulated Experiences
Trains a G1 humanoid in simulation to see an object, move to it, pick it up, and bring it back.
paper preview
13arXiv
EATR-Stereo: Embodiment-Aware Routing of Paired Stereo Evidence for Humanoid Vision-Language-Action Control
Routes stereo camera evidence using humanoid body state so a VLA can keep acting when views are occluded.
project clip
14arXiv
Video2DoorTraversal: Push Door Traversal via Simulated Door Twins
Turns a single real door video into a simulated door twin, then trains a wheel-legged robot to push through it.
paper preview
15arXiv
DECOWAM: Decoupled Whole-Body World-Action Model for Legged Mobile Manipulation
Separates base motion, arm motion, and camera motion so a legged mobile manipulator can learn world-action dynamics.
3 papers
Whole-Body Control
MPC, impedance, safety layers, and whole-body coordination.
paper preview
16arXiv
Throwing a Tight Spiral American Football by a Humanoid Robot
Coordinates a G1 humanoid's body, wrist, and fingers so it can throw a tight spiral football.
paper preview
17arXiv
Real-Time Control-Constrained DDP for Underactuated Balancing of Legged Robots
Runs constrained online planning fast enough to balance and move underactuated legged robots in real time.
paper preview
18arXiv
Hybrid Feedback Sampling for Sample-Efficient Model Predictive Control
Makes sampling-based MPC more stable and sample-efficient for contact-rich robot control.
3 papers
Training Infrastructure
Benchmarks, datasets, rewards, simulators, and tooling.
project clip
19arXiv
HumanoidVLN: A Physics-Grounded Simulator and Benchmark for Vision-Language Navigation Across Diverse Humanoid Embodiments
Benchmarks humanoid navigation by testing how different robot bodies actually walk, turn, and fail in a physics simulator.
paper preview
20arXiv
Coverage Aware Active Evaluation for Failure Discovery with Paired Systems
Finds robot failures faster by choosing the most useful real-system tests instead of testing every scenario.
paper preview
21arXiv
SCAPE: Scenario-Conditioned Simulation-Augmented Policy Evaluation
Evaluates robot policies with corrected simulation predictions and uncertainty from limited real-world data.