Learning Multi-gait Quadruped Locomotion for Interactive Media Art Exhibition | |
Legged robots are increasingly expanding beyond their traditional role in robust terrain navigation, finding new applications in the media art and entertainment sector as physical characters and interactive installation mediums. Quadruped robots, in particular, stand out for their ability to convey emotions and moods to audiences through their distinctive stepping patterns. In this context, we aim to explore the potential of using quadruped robots in interactive media art by developing a reinforcement learning (RL)-based locomotion controller capable of producing diverse and expressive gait and movement patterns. As part of an ongoing collaboration between the artistic studio AATB and the Computational Robotics Lab (CRL) at ETH Zurich, which focuses on developing and refining various subsystems of quadruped robots—including perception, navigation, communication, and locomotion for media art applications—this project centers on locomotion control and behavior synthesis. The primary objective of the project is to replace the built-in locomotion controller of Unitree Go1/Go2 robots with RL-based controllers, enabling more expressive gait patterns, movement styles, and enhanced interaction with the environment and audiences. The project is regularly exhibited internationally in various art venues, so we emphasize that the implementation aspect will be a key evaluation criterion. Ultimately, we expect the student to implement and successfully deploy their controllers for use in a live setting. More specifically, the student is expected to deliver an RL-based legged locomotion controller for the Unitree Go1/Go2 quadruped robots, capable of exhibiting diverse gaits and movement patterns. It is crucial that the entire pipeline runs robustly to ensure reliability in exhibitions. | |
Advanced Co-Design Framework for Legged Robots | |
This project seeks to advance the field of legged robotics by creating a versatile and accessible co-design framework that integrates mechanical design and control optimization. | |
Applying differentiable simulation for policy optimization in robotics | |
The project aims to develop new methods to use gradient information from differentiable simulations for policy optimization. |
Powered by SiROP - the academic career network