Inside China’s Largest Robot Exhibition | World Robot Conference 2025
Automatic Addison presents the World Robot Conference 2025 in Beijing, showing robots for service, research, industry, movement, and performance.
A Galaxea presenter discusses how MIT and NVIDIA’s Flash research enables Galaxea R1 Lite, paired with π₀.₅, to play ping-pong in real time.
The presenter says vision-language-action models have struggled to respond quickly and move smoothly during rapid tasks. Some demonstrations were sped up five to ten times to appear fluid.
Flash runs inference for the next action while the robot carries out its current one. It estimates the robot’s future state and uses that estimate with current visual input to choose the next action. The presenter says this approach leaves the original model architecture intact and adds no runtime overhead.
The presenter reports that Flash delivers about 2.03 times the inference and control speed, while reaction latency falls to about 5.7% of its original level. These changes support continuous movement in fast-changing settings such as ping-pong and whack-a-mole.
The research team paired Galaxea R1 Lite with π₀.₅ and reached an inference frequency above 30 hertz. The presenter says this setup enables real-time interaction in ping-pong and other fast tasks.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Automatic Addison presents the World Robot Conference 2025 in Beijing, showing robots for service, research, industry, movement, and performance.