ETH Zurich’s robotics researchers have introduced amazing achievements in robotics with the unveiling of ANYmal, a four-legged robot equipped with advanced AI modules. These modules empower ANYmal to navigate intricate terrains independently, marking a significant advancement in quadrupedal robotics.
The ANYmal robot integrates a sophisticated vision module, utilizing six Intel Realsense depth cameras and a Velodyne LiDAR powered by a Nvidia Jetson Orin controller. Overcoming the limitations of traditional laser and infrared scanning, which often fail to provide comprehensive spatial information, this module reconstructs the environment in 3D from the acquired scans. The robot comprehensively understands its surroundings by employing advanced imaging techniques, enabling seamless navigation through diverse obstacles.
Movement module: Dynamic adaptation to varied challenges
Comprising five distinct actions, including climbing, crouching, jumping, and walking, the movement module undergoes rigorous training to master each skill under varying difficulty levels. Through simulated scenarios, the virtual ANYmal robot is trained to execute intricate maneuvers such as crouching beneath progressively lower obstacles or leaping across widening gaps between platforms. This rigorous training regimen ensures the robot’s ability to dynamically adapt to diverse terrains and obstacles encountered in real-world environments.
The navigation module is the central intelligence unit, synthesizing data from the vision module with positional information and movement requirements to chart optimal paths through complex courses. Trained on an extensive dataset comprising 3000 test courses, the module achieves an impressive success rate of over 96% in navigating challenging terrains. The ANYmal robot demonstrates unparalleled precision in traversing intricate environments with remarkable efficiency by seamlessly integrating perception, spatial awareness, and motor skills.
Real-world deployment: Performance beyond simulation
Upon completion of simulation training, the AI software brain is seamlessly integrated into the physical ANYmal D robot, equipped with powerful hardware including two sets of Intel i7 CPU, 8GB RAM, and 240 GB SSD. The ANYmal robot showcases its exceptional capabilities in real-world scenarios by effortlessly surmounting obstacles and traversing challenging terrains with unparalleled agility. Videos capturing the robot’s performance underscore its remarkable ability to overcome obstacles impeding conventional robotic systems, heralding a new era in autonomous mobility.
With its advancements in autonomous navigation, the ETH Zurich research team has propelled the field of robotics into new frontiers. By harnessing the power of highly trained AI modules, the ANYmal robot demonstrates unparalleled versatility and adaptability in navigating diverse environments independently. As robotics continues to evolve, innovations such as these promise to revolutionize various industries, from search and rescue missions to exploration and beyond.
Original story from : https://www.anybotics.com/anymal-technical-specifications.pdf
