From teleoperation to onboard vision, the evolution of robots that see the ball
Demos of humanoids handling incoming balls can be graded by where the sensing sits. Tesla Optimus's tennis-ball catch in November 2024 was teleoperation by a person watching a camera feed, and what it proved was hand hardware whose degrees of freedom had risen from 11 to 22 (25 including the wrist), not AI autonomy. In the baseball-catching and batting robots of around 2025 and China's tennis robot in 2026, infrared-marker-based motion capture (120–240 Hz) set up around the playing area tracks the positions of the ball and the robot and passes them on, while the robot handles only decisions and whole-body movement. The most advanced stage is juggling, in which a robot repeatedly sees, catches and throws using only its own cameras, with no external equipment; its range of action is narrow, but it is fully autonomous from perception to movement. The forecast that 20–30-shot rallies and human-level movement speed are hardware problems achievable within 1–2 years is a personal estimate, and the names of robots and companies and the detailed figures need to be re-checked against the original demo materials.
Key points
- The Optimus catch was teleoperated; what it proved was hardware, hand degrees of freedom 11 → 22
- For the tennis and baseball robots, external motion capture (120–240 Hz) computes the trajectories
- A robot juggling repeatedly with onboard vision alone is the current peak of full autonomy
- Pro-level rallies are a hardware problem; dropping motion capture is forecast within 1 year (an estimate)
This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.