Adaptive Motion Planning Brings Athletic Moves to Humanoid Robots
Researchers have developed a system that helps humanoid robots mimic pro athletes’ movements in ball sports by separating the high-level planning from the precise motion tracking. Their method uses broadcast video and motion-capture data to teach robots how to serve and rally, with special training to handle the jump from simulation to real courts. The technique was tested on two humanoid robots performing real-world serves, even without motion-capture setups.
- System uses video and motion-capture for training
- Planner and tracker roles separated for stability
- Adaptive speed control improves sim-to-real performance
- Tested with in-the-wild tennis serves on two robots
- Motion tracking works without external sensors
Sources covering this
More in Future Tech
Antarctic sea ice maximum hits third lowest on record
NASA and the US National Snow and Ice Data Center reported that Antarctic sea ice reached its annual maximum of about 17.59 million…
Oxford team simulates quantum effect on hybrid computer
Oxford physicists have used a hybrid quantum computer—one that combines qubits with quantum oscillators—to simulate the Aharonov–Bohm…
Ghost of Yotei PS5 emulated but at 1-2 FPS on PC
A developer has managed to get Ghost of Yotei—the upcoming PlayStation 5 samurai game—running on the experimental SharpEmu emulator for PC.