The field of robotics has reached new heights, with ANYmal, the quadruped robot, mastering novel abilities. Initially, ANYmal had limited traversal skills but proved its adaptability on the rocky terrain of Swiss hiking trails. Recently, research at ETH Zurich has shown that ANYmal can now perform parkour and navigate complex terrains, like those found on construction sites or disaster-ridden areas.
These unprecedented skills have been attributed to the work of two teams, both supervised by Professor Marco Hutter of the Department of Mechanical and Process Engineering at ETH Zurich. One team is significantly led by an ETH doctoral student and parkour enthusiast, Nikita Rudin. He claims that despite doubts about the potential advancements of legged robots, he was confident that there was still room for innovation leveraging robotic mechanics.
Aligning his parkour experiences with his studies, Rudin embarked on a mission to extend ANYmal's capabilities using machine learning. His endeavours have led to impressive feats by ANYmal—scaling obstacles and executing dynamic maneuvers with precision. Rudin employed a trial and error approach for ANYmal's learning framework, akin to child-like learning. Given an obstacle, ANYmal uses its camera and artificial neural networks to identify and subsequently tackle the challenge based on its past learnings.
Rudin posits that while the individual skill growth of ANYmal might have peaked, there's still considerable scope for improvement. He suggested the implementation of real-world applications like manipulating ANYmal to negotiate challenging terrains, akin to those found in disaster areas.
The other project team, led by Rudin's colleague and ETH Doctoral student, Fabian Jenelten, aimed to prepare ANYmal for real-world applications. However, Jenelten differed in his approach to this by not solely relying on machine learning. Instead, he combined machine learning with a control engineering technique called model-based control. This fusion of technologies facilitated a smooth and more precise teaching method for ANYmal's maneuverability, such as recognizing and surmounting gaps in piles of rubble or other intricate terrains.
Jenelten claims that the combination of machine learning with model-based control allows for the optimal use of ANYmal. The quadruped robot is now more equipped to navigate slippery or unstable surfaces. It'll soon be operational in hazardous sites such as disaster-struck areas or construction zones—areas of high risk for human beings.
Disclaimer: The above article was written with the assistance of AI. The original sources can be found on ScienceDaily.
