EU Horizon 2020 funded researchers have developed a new control framework for the four-legged, dog-sized autonomous robot ANYmal, significantly enhancing its ability to traverse challenging terrains. This advancement comes as a breakthrough in the field of robotics, particularly in enhancing the capabilities of robots in real-world applications such as search-and-rescue operations.
The newly developed framework allows ANYmal to effectively navigate across various difficult surfaces, including staircases, foam blocks, and rocky grounds. This improvement is attributed to the integration of a deep tracking control framework that combines trajectory optimization with refined reinforcement learning (RL). This hybrid approach enables the robot to select more appropriate footholds and adjust its joint positions more effectively, leading to a notable enhancement in its stability and recovery from slips.
During testing, the researchers focused on simulating challenging environments that mimic real-life scenarios, such as natural disaster sites with potential debris and construction sites with slippery surfaces. The team conducted extensive simulations, training over 4,000 robot instances across more than 76,000 square meters of diverse terrain. This rigorous preparation was essential before deploying the control framework in actual robots navigating an obstacle course.
The performance of ANYmal under this new framework has shown significant improvement over previous iterations. As noted by Fabian Jenelten and colleagues, the robot outperformed the RL baseline controller on every tested terrain, showcasing its enhanced adaptability and resilience. This development marks a significant step forward in the field of autonomous robotics, particularly in enhancing their utility in complex and unpredictable environments, such as those encountered in search-and-rescue missions.
Phot credit: Fabian Jenelten, Junzhe He, Farbod Farshidian, and Marco Hutter
