A robot dog that can pick up a ball and hurl it across a field sounds like something a Pixar movie, but researchers at ETH Zurich have turned this concept into reality. Their creation, a quadruped named ANYmal, doesn’t just walk or run—it manipulates objects with a precision that rivals human dexterity.
Most robots are trained for repetitive jobs in tidy factories, but ANYmal needed to handle the chaos of uneven ground, shifting objects, and the tricky dance of grabbing and throwing something on the move. The team at ETH Zurich’s Robotic Systems Lab, led by Fabian Jenelten, gave ANYmal a sleek, custom arm mounted on its back. With a simple gripper, this arm turned the robot into a versatile maestro. “We wanted to see how far we could push a legged robot,” Jenelten says, his excitement palpable.
- 🎁Ideal Gift for Kids & Teens: Celebrate child’s growing skills and important milestones with this 5-in-1 Programmable robot set. Whether for...
- ✨STEM Educational Toys: The robot set for kids ages 8+ combines the fun of STEM learning. It encourages hands-on learning and early programming as...
- 📱Flexible Dual Control Modes: Control the Robotic kit with the intuitive app (Bluetooth) or remote. Enjoy fun features like basic programming...
To achieve this, the team leaned on reinforcement learning, and set ANYmal loose in a virtual world, letting it practice millions of throws without risking a smashed apple or a dented robot. This digital sandbox wasn’t just for fun—it mirrored the real world’s physics, from gravity to the way a ball skids on grass. Using a trick called sim-to-real transfer, they made sure ANYmal’s virtual skills worked just as well in reality. “Simulation’s a game-changer,” Jenelten explains. “You can try endless scenarios, but the real win is making those lessons hold up when the robot steps outside.”
The arm itself moves with a lifelike fluidity, thanks to its six degrees of freedom. Its gripper, simple as it looks, can handle anything from an apple to a tennis ball. The team taught ANYmal to scoop objects off the ground, adjust its stance to stay balanced, and fling them with uncanny accuracy. In one video, it lobs a ball to nail a target meters away, a move that’s less about brute force and more about finesse. “Robustness was our focus,” says researcher Joonho Lee. “This robot has to deal with different weights, shapes, even a gust of wind, and still get it right.”
When it came time for real-world tests, ANYmal didn’t disappoint. The team scattered balls, bottles, and fruit across grass, gravel, and concrete, and the robot handled it all, tossing objects to hit targets up to five meters away. They even threw curveballs—literally—like windy days or bumpy terrain, and ANYmal adjusted on the fly. Unlike older robots that need every step scripted, this one’s learning model lets it adapt to life’s unpredictability. “The real world’s messy,” Jenelten says with a grin. “Most robots trip over that, but ANYmal? It thrives.”
[Source]
