Machine learning is promising

Since ETH published their learning papers on Science Robotics. Learning is promising for any robots.

[1] Hwangbo, Jemin, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter. "Learning agile and dynamic motor skills for legged robots." Science Robotics 4, no. 26 (2019).

[2] Lee, Joonho, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, and Marco Hutter. "Learning quadrupedal locomotion over challenging terrain." Science robotics 5, no. 47 (2020).

Tool-kits

RaiSim ANYmal team developed this physical engine for simulation. They compared every physical engine from different aspects. Please go to this web to see the detail. There conclusion is RaiSim is the best for machine learning for legged robots.

Environment set-up

PyTorch and TensorFlow are the main neural network libraries to play the architecture of NN.
Open AI gym is a good tool to set up everything.
Some guys also released their wrappers for connecting different simulators and gym, for example, gym-gazebo.

Google Sites

Report abuse