Robot dog learns to open doors with its leg. Now there is nowhere to hide

In Switzerland, a team from ETH Zurich has pushed boundaries in robotics by teaching a robotic dog, the ANYmal, to perform tasks using its legs rather than arms. Led by Philip Arm, the researchers employed a machine learning model to instruct the robot in activities such as opening doors, pressing buttons, and carrying objects like a rucksack. The innovation aims to optimize the robot’s functionality for scenarios where dexterity is less critical, such as in space exploration where weight constraints are paramount.

By leveraging machine learning, the team trained the ANYmal to balance on three legs while utilizing the fourth for tasks, enhancing the robot’s agility and reducing its overall weight. Additionally, the researchers at the Robotics Systems Lab are exploring lifelike agility inspired by animal locomotion, including utilizing the robot’s tail for stability during various movements.

The ANYmal robot, developed by ANYbotics, is designed as a rugged, autonomous inspection solution equipped with advanced sensors. It is adaptable to diverse environments, including harsh conditions like rain, snow, wind, waterlogged areas, and dusty surroundings. The team at ETH Zurich has integrated the ANYmal into their projects to enhance inspection and manipulation tasks.

In summary, the groundbreaking work in Switzerland showcases the potential for machine learning to enhance robotic capabilities, illustrating a shift towards leg-based functionality in robotic systems. This innovation could have significant implications for fields requiring agile, lightweight robotic solutions, such as space exploration and industrial inspections.