Kai Yuan (cohort student representative)

Research project title: 
Control and Learning of Versatile Legged Mobility on Complex Terrain
Principal goal for project: 
The goal of my research is to explore and implement control and learning schemes to achieve locomotion over complex terrain for legged robots (bipeds and quadrupeds). In the scope of my research a design control framework will be developed that is independent of the amounts of legs and is able to guide the robot over obstructed terrain (e.g. a box of variable height). Furthermore, it will be investigated how and to which extend Machine Learning and AI concepts can solve problems that classical control methods struggle with. First, a proof of concept is delivered via simulations and later implemented on the robots, such as Valkyrie and ANYmal.
Research project: 

The biggest advantage legged locomotion has over wheeled locomotion is its potential ability to locomote over obstructed, elevated, rough, and abrupt terrains. In the domain of bipedal locomotion, the method of purely bipedal realization, i.e. without the usage of arms for additional stability, is predominant. This is partly due to the nature of the environment where humanoids interact with. In these environments, which are often flat grounded and don't offer any surfaces that can be used to place the hands onto, bipedal locomotion performs well, and the necessity of multi-contact locomotion does not arise or is not possible. In these kinds of environments legged locomotion do not have any advantages over wheeled locomotion, which leads to the conclusion, that more focus should be put onto environments and terrains on which wheeled locomotion fails to perform. Examples for such terrains and environments are highly cluttered with large obstacles (e.g., disaster zones, terrain in nature, stairs). In these kinds of environments versatile motion skills (using the arms for additional support) offer additional robustness: Stair-climbing gets significantly easier by using the rail, locomoting over a high obstacle is easier when the hands are used instead of stepping over it, locomoting over cluttered ground becomes easier when a nearby wall can be used to stabilize our gait. This project aims to develop a suitable optimization framework for autonomous locomotion over arbitrary cluttered and obstructed terrain without prior restrictions on contact points, motion patterns, etc, and the implementation of the framework on Valkyrie.

 

About me: 

I am a PhD student at the Edinburgh Centre for Robotics under the supervision of Dr. Zhibin Li and Dr. Timothy Hospedales. My main interest lies in the control and specifically optimisation of robots and the application of Machine Learning to further enhance the autonomy and intelligence of robots. I hold a BSc and MSc in Engineering Cybernetics from the University of Stuttgart (Stuttgart, Germany), and an MSc by Research with Distinction in Robotics and Autonomous Systems from The University of Edinburgh. I previously worked as Researcher at Tsinghua University (Beijing, China), Senior Algorithm Engineer at UBTECH Robotics (Beijing, China), and as Corporate Research Intern at Robert Bosch GmbH (Renningen, Germany).

 
Publications: 
  • C. Yang, K. Yuan, W. Merkt, T. Komura, S. Vijayakumar, and Z. Li, “Deep Reinforcement Learning of Locomotion Skills for the Humanoid Valkyrie” , International Conference on Humanoid Robots (Humanoids), 2018
  • K. Yuan and Z. Li, “An Improved Formulation for Model Predictive Control of Legged Robots for Gait Planning and Feedback Control”, International Conference on Intelligent Robots and Systems (IROS), 2018
  • W. Hu, I. Chatzinikolaidis, K. Yuan, and Z. Li, “Comparison Study of Nonlinear Optimization of Step Durations and Foot Placement for Dynamic Walking” , International Conference on Robotics and Automation (ICRA), 2018
Student type: 
Current student