Manufacturer of the Unitree G1 humanoid robot used to validate ViBe.
“We evaluate on four perceptive control tasks on a Unitree G1.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.
“MuJoCo Warp [16] handles the robot's whole-body and rigid contact dynamics, while the MPM solver computes contact forces on deformable sand.”
Source→“The student has no decoder of its own and reuses the teacher's fixed decoder during training... steering the latent representation specifically toward terrain material properties.”
Source→“It uses a 'teacher-student' framework where the teacher trains with privileged information (ground-truth terrain stiffness) compressed into a compact latent representation via a variational autoencoder (VAE).”
Source→“The paper's core technical contribution is the integration of a three-dimensional resistive force theory (3D RFT) contact model into a reinforcement learning (RL) training pipeline.”
Source→“CLAP demonstrates few-shot adaptation to the G1 humanoid's 26-dimensional action space (7-DoF per arm, 6-DoF per hand). The adapted model achieves PSNR 15.151 on G1 video prediction”
Source→“The biggest opportunity for humanoids may be one that Unitree has barely tapped: automating the factory floor.”
Source→“Backflipping robot maker Unitree raised $904 million in its Shanghai IPO, yet the biggest opportunity for humanoids may be one that Unitree has barely tapped: automating the factory floor.”
Source→“On a real Unitree Go2 robot, our system maintains stable locomotion under severe instances of the evaluated changes, including a fully locked leg and a 5 kg payload, where non-adaptive methods fail" (Abstract).”
Source→“We use the Unitree Go2 quadruped as the robot platform. The robot's nominal mass is 15.7 kg, including the battery, and it has 12 actuated degrees of freedom" (Section IV-A).”
Source→“The paper demonstrates that the mathematical safety constraints used to train a robot must be explicitly matched to the quality of the robot's sensors.”
Source→“We therefore deploy a lightweight Link-CBF policy zero-shot on the Unitree G1 in the real world, where it tolerates imperfect perception, succeeds on 95% of throws”
Source→“Unitree is at least a large percentage open source. And it looks like they have just had such a booming industry and so much deployment within China.”
Source→“Unitree Won Shanghai STAR Market IPO approval in June 2026 at ~42B yuan valuation, 73 days from application to approval.”
“At least companies like Unitree and Zhiyuan have already gotten mass production running — the supply chain is also maturing as the industry develops.”
Source→“evaluated on a supermarket chip-restocking task using a Unitree G1-Edu humanoid robot”
Source→“The Go1 is noted as "the most agile of the four platforms and executes wide turns without re-planning." The G1 experienced calibration drift of ~2cm after 30 minutes, causing overly wide turns at doorways.”
Source→“Qwen3.5-27B runs for VA in simulation and on the wheeled platform; Qwen3.5-9B-Q4 runs locally on the Unitree G1 and Go1's Jetson Orin NX.”
Source→“Uni-LaViRA: Language-Vision-Robot Actions Translation for Unified Embodied Navigation”
Source→“Preliminary results of our method applied to different robots—such as ANYmal, Go1, and KAIST HOUND in bipedal mode—are shown in movie S4 (Section 3)”
Source→“Integrating the manipulator into the calf places this reach close to the ground: the arm base sits 0.18 m above the floor against 0.36 m for a trunk mount, so a trunk-mounted arm would need twice the reach to grasp the same floor-level object.”
Source→“This paper integrates a manipulator with a prismatic slider, two revolute joints, and a gripper into each front calf of a Unitree Go2.”
Source→“The real-world evaluation shows high success rates for navigation (e.g., 20/20 for 'Walk To' in the lab scene)”
Source→“We evaluate on the physical Unitree G1 performing navigation and single-object transport”
Source→“CoorDex argues that with the right latent space structure, high-DoF hands are not only trainable but necessary for continuous loco-manipulation”
Source→“The G1 is the physical platform used for all experiments and real-world demonstrations. "Our experiments use a 29-DoF Unitree G1 humanoid"”
Source→“CoorDex enables a Unitree G1 humanoid to perform tasks like grasping a bottle, opening a fridge, and picking up a cube without stopping”
Source→“OpenHLM: An Empirical Recipe for Whole-Body Humanoid Loco-Manipulation”
Source→“All our loco-manipulation tasks are carried out by a Unitree G1 robot...”
Source→“π0.5 reaches 91% average task progress, PaliGemma drops to 60%, and random initialization collapses to 42%”
Source→“Someone did a translation, the Unitree, there's a humanoid company, Unitree, that's going public in China, and it's doing 500 million. It's a real company. It's profitable”
Source→“China's 160-Company 'Robot Bubble' Is a Supply Chain Advantage, Not a Problem”
Source→