HK university robot performs basketball layup using AI

Researchers at the Hong Kong University of Science and Technology (HKUST) have demonstrated a 1.3-meter-tall humanoid robot, Unitree G1, performing a real-world basketball layup.

This is the first time a robot has executed basketball moves in a real environment.

The project builds on HKUST’s earlier work using imitation learning and physics-based control to teach robots human-like skills.

The robot’s abilities are powered by SkillMimic-V2, a new method for learning complex, generalizable actions from limited and imperfect demonstration data.

SkillMimic-V2 was accepted for SIGGRAPH 2025 and uses techniques such as Stitched Trajectory Graphs and Adaptive Trajectory Sampling to improve the robot’s skill transition and robustness.

🔗 Source: Pandaily

🧠 Food for thought

Implications, context, and why it matters.

The demo’s success likely depends on controlled conditions that aren’t fully disclosed

That 91.5% layup rate in simulation sounds strong, yet the team did not share real-world rate or attempt count 1.
Key setup details remain unclear such as onboard perception versus motion capture or a pre-grasped ball or a fixed start pose 1.
SkillMimic-V2 handles noisy sparse data and recovers via State Transition Fields, yet the jump from 91.5% in simulation to real use stays murky 2.
First in a real environment sounds meaningful, yet the lack of trial counts and failure modes blocks any call on reproducibility 1.

Open-source robotics developers can build skill libraries if SkillMimic-V2 code is released permissively

The SkillMimic repo already ships code, pre-trained models, and datasets in public form, so SkillMimic-V2 may do the same 3.
With a permissive license, outside developers could offer training-as-a-service that lets smaller robotics teams learn interaction skills without standing up full simulation stacks.
Cloud simulators may bundle SkillMimic-V2 into ready setups, so startups can prototype and validate behaviors before touching hardware.
Tool vendors may ship eval suites plus marketplaces where teams share or sell pre-trained basketball, household, or industrial skills from this framework.