At ETH Zurich, researchers have enhanced the capabilities of ANYmal, a quadrupedal robot, enabling it to perform parkour and navigate through challenging terrains, such as those encountered in building sites and disaster zones. Previously competent in traversing the terrain of Swiss hiking trails, ANYmal has now been trained in parkour, an athletic activity that involves overcoming obstacles often found in urban settings.
This development was led by two teams working under ETH Professor Marco Hutter from the Department of Mechanical and Process Engineering. Nikita Rudin, an ETH doctoral student and part of one of these teams, utilized his parkour experience to extend the mechanical capabilities of ANYmal. He employed machine learning to teach the robot to climb over obstacles and perform dynamic jumps. The learning method used for ANYmal involved trial and error, similar to learning patterns in humans. The robot utilizes its camera and artificial neural network to evaluate obstacles and select appropriate movements from its training.
Fabian Jenelten, another ETH doctoral student, led the second team with a focus on preparing ANYmal for operation in environments like disaster sites with uneven terrain. Their approach combined machine learning with model-based control, a standard method in control engineering. This strategy allowed ANYmal to accurately maneuver through complex terrain, such as navigating through rubble. Machine learning aids the robot in adapting to unexpected situations.
Due to these enhancements, ANYmal has improved its stability on surfaces that are slippery or uneven. The robot is set to be deployed in areas such as construction sites and disaster zones, performing tasks like inspecting damaged buildings, which are potentially hazardous for humans. This advancement in ANYmal’s capabilities is part of a wider trend in robotics where integrating new technologies with established methods results in robots better equipped to handle diverse and challenging environments.
