Skip to content
QbitAI· 十三·· 5 d agoPickAI score40

TARS Team Wins First Embodied Large Model Challenge by Collecting Custom Data in 48 Hours

真勇啊!00后团队参加具身挑战赛,仅48小时竟敢从零手搓数据,还夺冠了!

AI summary

In the finals of Ant Lingbo's inaugural embodied large model challenge, the TARS team—with an average age of 25—collected custom data and completed three robotic arm manipulation tasks within 48 hours, scoring 82 points to claim first place and 80,000 yuan in prize money. Diverging from the official 210 grasping trajectories, TARS refined the grasping strategy (switching from handle to edge) and recollected data, overtaking the second-place EVA team in the final moments.

Why it matters

The finals demonstrated a typical iteration cycle for embodied AI: how data collection strategy, model fine-tuning, and real-world engineering adaptation trade off within 48 hours, while different teams' solutions provide concrete case studies for the industry.

Source: QbitAI · qbitai.com