Blind Quadruped Robot Locomotion Across Different Terrains скачать в хорошем качестве

Blind Quadruped Robot Locomotion Across Different Terrains

In this project, our team aimed to enable a quadruped robot to navigate various terrains without relying on visual input. To achieve this, we began by studying a research paper that utilized Beta Variational Autoencoders to help robots interpret their surrounding environment using past proprioceptive data. Based on the paper's idea, we developed our own Beta Variational Autoencoder code. We then implemented it in Isaac Gym and trained the Unitree Aliengo robot to walk on various terrains using only proprioceptive data. Once the policy and encoder were successfully trained, we transferred them to the actual Aliengo robot for testing and real-world application. The name of the referenced paper is: "DreamWaQ: Learning Robust Quadrupedal Locomotion With Implicit Terrain Imagination via Deep Reinforcement Learning" https://arxiv.org/abs/2301.10602