Tsinghua and ByteDance publish vision-driven soccer robot framework in Science Robotics
jiqizhixin · x · 2026-08-28
Tsinghua University, ByteDance Seed, and China Agricultural University published a vision-driven reactive soccer framework for humanoid robots in Science Robotics. The robot relies solely on onboard vision to find, chase, and shoot the ball, replacing external motion capture or pre-scripted behaviors.
Key Results:
- Accuracy: An internal state estimator cuts ball position error from 0.344m to 0.186m (46% reduction), allowing tracking even when vision is blocked.
- Speed: Time from start to ball contact is 1.5s, a 64% speedup over traditional rule-based systems.
- Reliability: Front-field shooting success reaches 80-90%, back-field 60-70%, with zero falls during testing. Success remains high even after 0.3s of vision interruption.
More from Embodied
- XSquare Robot unveils WALL-SS: long-horizon world model with 0.93 sim-real correlation — chris_j_paxton · 2026-08-28
- Crazy moment from China's AI robot 100m race — jjvincent · 2026-08-28
- ROS Robots Collide on Network Due to Missing Namespaces — chrismatthieu · 2026-08-28
- OpenRoboto brings real-robot validation, ending simulation-only judging — const_reborn · 2026-08-28
- Boston Dynamics reveals how 'robot surgeons' fix Atlas — mattbeane · 2026-08-28
- CGS-SLAM: Collaborative Gaussian Splatting SLAM for Multi-Agent Reconstruction — zhenjun_zhao · 2026-08-28