Interview with Grit Ventures Academic Council Member, Dr. Changliu Liu, Carnegie Mellon University
Dr Changliu Liu is an assistant professor at the Robotics Institute at CMU where she leads the Intelligent Control Lab. Dr. Liu is also a member of Grit’s Academic Council.
Changliu grew up in Shanghai China. When she was young her parents encouraged her to study science. She wasn’t interested in chemistry or biology, but fell in love with physics and engineering and actually building robots.
I was intrigued by the potential to build a machine that could physically look like humans and do things that humans can do. I knew that if we wanted to make robots smart, we would need to enable Artificial Intelligence.
After completing her undergraduate degree in Beijing, She obtained her Ph.D from Berkeley in 2017, where she worked in Mechanical Systems & Control Lab. Prior to joining CMU, Dr. Liu was a postdoc at Stanford Intelligent Systems Laboratory.
What brought you to CMU?
My primary research focus is on the design and verification of intelligent systems that work with people, with application to manufacturing and transportation. CMU has a truly unique Robotics Institute. It’s incomparable. We collaborate across multiple disciplines. Conventionally robotics resides in Computer Science or Mechanical Engineering, but CMU has a standalone department. Robotics is cross-disciplinary. As you are solving problems, you need to bring these disciplines together.
Tell me more about your research?
I’m just starting to build my Lab. I have 7 PhD students, 3 master students, some visitors, and some staff members. My hope is that half of my PhD students will go into academia and half will go into industry. We have two big collaborative projects with industry: One is with Siemens and Yaskawa. In that project, we are applying real time robot motion planning and control techniques to solve a very specific manufacturing problems, to polish the interior of a big pipe-like workpiece.
My other project is to assemble a very accurate and flexible robotic platform that can do delicate PCB assembly. We are using a lot of real time motion planning and control skills and real time learning skills in order for the robot to adapt to the variety of pieces and types of defects. At a high level you can classify all of these projects as real time planning and control under uncertainty. But the real time requirements and sources of uncertainty are different.
In the assembling use case, we have pins on IC chips that can be bent. The robot learns from the failed insertion and then adjusts the strategy.
Our work on Advanced Robotics for Manufacturing
Do you see a trend for more commercially sponsored research vs. government sponsored?
There is clear cut interest from both. The government is interested in future manufacturing techniques and human/robotic interaction, for example, human-robot collaboration in electronic assembly. Commercial partners are less interested in direct collaboration between humans and robots, though they may share the workspace. They would prefer to focus on robotic automation first, that is to expand robots’ skillsets and improve their dexterity first. It’s not clear when we’ll see commercialization of close human-robot collaboration in e.g. assembly lines but it will certainly create new use case and market opportunities in the future.
What are some other interesting areas of research at CMU?
CMU is known for Autonomous Vehicles and vehicles on the moon. We launch robots in the wild through our Big Field Robotics Center for example. CMU is also developing interesting applications using AI to enable robots to serve people in work and home environment which are inherently unstructured. In contrast, manufacturing environment are very structured but at home, the robot needs to be more flexible and needs to make decisions faster.
When you think about commercializing technology from the labs, what resources are most helpful?
Find a problem is not the major issue but launching a business is challenging for us academics. We are good at writing proposals, writing papers, and managing research but it’s challenging to bridge product development to meet a customer’s needs. Researchers also need help building the business model. We need to be educated on all these issues to take launch a company.
CMU provides resources on the customer discovery side. The School of Computer Science has a business liaison office that does a lot of outreach to industry partners. Once there is a specific request from a partner, they identify faculty members with relevant research. Before Covid, I used to get emails once per month asking if I’d like to meet with representatives from a specific company. On the commercial side, business development and technical people come and present their use cases. National Labs sends technical people. This is a tremendous sourcing of funding for new projects.
What do you imagine 10 years from now in terms of robotic applications?
Most likely wide spread deployments of robots will be in structured environments like factories. On the unstructured side, we’ll see moon vehicles. But in complex environments we won’t see broad deployments until we create more infrastructure for it to happen. We will have level 4 autonomy, buses, trucking.
Thank you Dr. Changliu Liu.
