| IN A NUTSHELL |
|
The development of robots that can autonomously interact with their environments marks a significant breakthrough in robotics. Among these innovations, the ANYmal robot, designed by ETH Zurich, stands out due to its remarkable capabilities. Equipped with four legs and cutting-edge technology, it navigates challenging terrains and now performs complex tasks like object launching. This article delves into ANYmal’s technical features, learning process, and real-world performance.
Technical Capabilities of ANYmal
ANYmal, developed by ETH Zurich’s robotics systems lab, is engineered to autonomously operate in some of the world’s toughest environments. Equipped with precision-controlled torque actuators, it can dynamically run, climb steep or uneven terrains, and maintain stability in complex contexts. This advanced technology enables ANYmal to optimally navigate its environment, carefully selecting the safest footholds.
Embedded laser sensors and cameras allow it to continuously perceive its surroundings, create detailed 3D maps, and accurately localize itself. This situational awareness is crucial for tasks like industrial inspection, disaster response, and exploration, where human access is challenging or hazardous.
Adding an Arm for Versatile Tasks
ANYmal stands out with its ability to operate in unpredictable terrains, unlike most robots designed for repetitive factory tasks. Under Fabian Jenelten’s leadership, the ETH Zurich team equipped ANYmal with a custom arm mounted on its back. This arm, featuring a simple yet effective gripper, transforms the quadruped into a versatile operator capable of grasping and launching objects while moving.
To maximize this arm’s capabilities, researchers employed reinforcement learning by releasing the robot in a virtual environment. In this digital sandbox, ANYmal could practice millions of throws without the risk of damaging objects or injuring itself. The aim was to test the limits of a legged robot’s capabilities, and the results were promising.
Simulation and Real-world Performance of ANYmal
The research team used the sim-to-real transfer technique to ensure ANYmal’s skills developed in simulation effectively apply in the real world. This method allowed the robot to learn adapting to unpredictable outdoor conditions, ensuring its acquired behaviors remain reliable outside the lab.
ANYmal’s arm, with its six degrees of freedom, provides realistic motion fluidity. Although the gripper seems simple, it can grasp a variety of objects, from apples to tennis balls. ANYmal was trained to pick up objects, adjust its position to maintain balance, and throw them with impressive precision. In real-world tests, the robot successfully threw objects at targets up to 16 feet away, even facing challenges like strong winds or rugged terrain.
Towards a More Adaptable Robotics
Fabian Jenelten emphasizes that, unlike older robots requiring precise action programming, ANYmal’s learning model allows it to adapt to unforeseen situations. This aspect is crucial in a rapidly changing world where adaptability is essential.
ANYmal’s ability to operate uninterruptedly in unpredictable environments makes it particularly valuable for demanding missions. It represents a new era of robotics, where autonomy and adaptability are at the heart of technological innovations. This raises a crucial question: how far can these advancements take us in our quest for even more autonomous and intelligent robots?
Did you like it? 4.6/5 (22)
