The programme targets an autonomous humanoid that runs its VLA model onboard a Korean-made AI chip.
LG Electronics oversees the humanoid system; Mobilint supplies the NPU; Nota leads VLA optimization using quantization, pruning, token compression and knowledge distillation.
NC AI’s work covers motion retargeting, training-data generation and digital-twin pipelines for humanoid learning.
OmniHand reached the finals in all eight dexterous-hand events and won seven gold medals, including block building, powder weighing and bean picking with tweezers.
X2 won the 100-metre obstacle race and A3 won the Tai Chi event; AgiBot says X2 competed in its mass-produced configuration without competition-specific hardware changes.
G2 contributed to five scenario-event golds across tasks including hotel service, library operation and emergency response.
The household task ran for up to 30 minutes and required navigation, object storage, clothing handling, door opening and interruption by an unexpected delivery task.
Galbot completed the food-service workflow in 6 minutes 55 seconds without an error, including voice-command handling, food pickup, microwave operation, drink preparation and delivery.
Its teams swept gold, silver and bronze in the retail event, where item locations, orders and stock conditions changed between runs.
Tiangong Ultra won the large-robot 100 metres in 8.64 seconds and the large-robot 400 metres in 38.15 seconds.
Tiangong also won the small-robot 400 metres in 45.66 seconds, while the platform recorded additional results in jumping, obstacles, weightlifting and scenario events.
The five-day Games concluded on August 26 after 1,301 contests involving 2,056 robots from 666 teams.