Unitree G1
Open-source AI for real-world robots. LeRobot is Hugging Face's library for teaching robots by example, enabling robots to learn tasks from a few demonstrations on hardware ranging from 3D-printed arms to full humanoids.
Primary use case: Teaching robots to perform tasks by demonstration for real-world robotic applications.
Height
132 cm
Weight
35 kg
DOF
23
Payload
3 kg
Battery
2 hrs
Price
$14K
Specifications
Physical
Height
132 cm
Weight
35 kg
Degrees of Freedom
23
Locomotion
bipedal
Manipulation
Hand Type
unknown
Payload
3 kg
Power & Autonomy
Battery Life
2 hrs
Charging Time
2 hrs
Autonomy Level
unknown
AI & Compute
AI Capabilities
Learning robot policies from human demonstrations, Imitation learning, Training vision-language-action models, Language-conditioned robot control, Simulation-based policy testing
Commercial
Price
$14K
Availability
Open-source and available now via `pip install lerobot`
Primary Use Case
Teaching robots to perform tasks by demonstration for real-world robotic applications.
Related news
Unitree's wild stock ride triggers warning from CEO
DeepSeek is leading a $400 million seed round for a stealth robot AI startup
BYD goes big in humanoid robots with 4,000 engineers, NVIDIA makes robots smarter, and much more
Renting a humanoid robot in China now costs $40 a day, down 80% in 1 year
WAM-bam-thank-you NVIDIA: Long-WAM gives robots vastly better skills
Apple's HALO Teaches a Unitree G1 to Navigate Unseen Rooms With Simulation-Generated Data
Trained on 300 hours of generated motion in simulated rooms, the onboard system completed 91% of 360 real-world trials without any teleoperation data.
RoboColosseum Launches Real-World Robot AI Benchmark With Unitree Support
The initiative offers more than 100 robots and 10,000 free evaluation hours, while initial G1 tests reveal how far robot policies remain from reliably completing tasks.
HALO Teaches Humanoids to Coordinate With Human Partners During Shared Carrying Tasks
HALO, a research project from the University of Texas at Arlington and Carnegie Mellon University, adjusts learning updates so humanoid policies better align with people during physically coupled tasks. A motion-capture-assisted Unitree G1 completed confined-space transport faster than PCGrad and scripted baselines, with no drops in that condition. The evidence remains preliminary because the experiments averaged only five trials, and the long-object halting task still had a 20% drop rate.