Research
Academic research and breakthroughs in humanoid robotics.
Research Areas
Locomotion & Balance
Advances in bipedal walking, running, and dynamic balance. Recent work includes terrain-conditioned motion priors, predictive style matching, and MPC-guided reinforcement learning.
120+ papers trackedManipulation & Dexterity
Progress in hand design, grasping, and fine motor skills. Vision-Language-Action (VLA) models are enabling general-purpose manipulation from language instructions.
95+ papers trackedAI & Cognition
Large language models, vision-language-action models, and reasoning for robot intelligence. Foundation models like GR0T and Helix are driving rapid progress.
180+ papers trackedHuman-Robot Interaction
Natural language communication, social robotics, and collaborative behaviors. Focus on safe and intuitive interaction in shared human environments.
65+ papers trackedActuators & Hardware
Electric motors, series elastic actuators, and novel mechanisms. The shift from hydraulic to electric systems is enabling more efficient and safer humanoids.
85+ papers trackedSimulation & Training
Sim-to-real transfer, reinforcement learning, and digital twin environments. Frameworks like QuadVerse align visual-physical reality for better transfer.
110+ papers trackedRecent Papers
Predictive Style Matching: Natural and Robust Humanoid Locomotion
A novel approach to generating natural humanoid locomotion by matching predicted motion styles to reference trajectories, improving robustness across diverse terrains.
T-GMP: Terrain-conditioned Generative Motion Priors for Versatile and Natural Humanoid Locomotion
Generative motion priors conditioned on terrain type enable humanoid robots to adapt their gait naturally to different surfaces without explicit terrain detection.
HANDOFF: Humanoid Agentic Task-Space Whole-Body Control via Distilled Complementary Teachers
A whole-body control framework for humanoids that distills multiple teacher policies into a single student policy capable of complex task-space manipulation.
MotionDisco: Motion Discovery for Extreme Humanoid Loco-Manipulation
Discovers novel motion primitives for humanoids performing extreme loco-manipulation tasks, combining locomotion and manipulation in challenging scenarios.
TAGA: Terrain-aware Active Gaze Learning for Generalizable Agile Humanoid Locomotion
Active gaze learning enables humanoid robots to proactively look at upcoming terrain, improving agility and generalization across unseen environments.
LadderMan: Learning Humanoid Perceptive Ladder Climbing
First successful demonstration of a humanoid robot climbing ladders using learned perceptive policies, expanding the range of environments humanoids can navigate.
Accelerating and Scaling MPC-Guided Reinforcement Learning for Humanoid Locomotion and Manipulation
Combines model predictive control with reinforcement learning to accelerate training of humanoid locomotion and manipulation policies while improving stability.
RhinoVLA Technical Report
A Vision-Language-Action model optimized for real-time deployment on edge hardware, enabling responsive humanoid manipulation without cloud dependency.
LARA: Latent Action Representation Alignment for Vision-Language-Action Models
Aligns latent action representations across vision, language, and action modalities to improve the performance of VLA models on complex manipulation tasks.
Shield-Loco: Shielding Locomotion Policies with Predictive Safety Filtering
A safety framework that wraps RL locomotion policies with predictive safety filters, preventing constraint violations while maintaining natural movement.