Description | Learning to walk and run | Jonathan W. Hurst, Oregon State University Abstract Machine learning is a powerful tool, but it’s not a black box. At the Dynamic Robotics Lab at Oregon State University, and also at Agility Robotics, we are working to understand natural legged mobility, and then use engineering tools including machine learning to capture the same physics that underlie animal and human mobility and manipulation, and implement on machines. By capturing the same physics, we aim to achieve the same performance. This talk will focus on examples of our research applying machine learning tools to a structured control system for legged locomotion on Cassie, our bipedal robot. We’ve recently achieved walking, running, skipping and jumping, as well as transitions between the gaits; and I’ll show several videos for the first time publicly. Biography Machine learning is a powerful tool, but it’s not a black box. At the Dynamic Robotics Lab at Oregon State University, and also at Agility Robotics, we are working to understand natural legged mobility, and then use engineering tools including machine learning to capture the same physics that underlie animal and human mobility and manipulation, and implement on machines. By capturing the same physics, we aim to achieve the same performance. This talk will focus on examples of our research applying machine learning tools to a structured control system for legged locomotion on Cassie, our bipedal robot. We’ve recently achieved walking, running, skipping and jumping, as well as transitions between the gaits; and I’ll show several videos for the first time publicly. |
---|