Q&A with the capstone winners of 2020 Fung Institute Mission Award
Jamie Chen (ME), Sonny Li (ME), David Tondreau (ME), and Mengyue Wang (ME)
The Fung Institute Mission Award is given to the capstone team that best exemplifies the mission of the institute: “Shaping generations of technical leaders to innovate across boundaries.” Finalists are nominated by Fung Instructors, and winners are chosen by Fung Institute staff based on the project brief.. This award is presented to the Adapting Humanoid Robots to Aid First Responders capstone team, Jamie Chen (ME), Sonny Li (ME), David Tondreau (ME), and Mengyue Wang (ME), advised by Koushil Sreenath and in collaboration with Hybrid Robotics Lab.
In an effort to mitigate first responders’ exposure to danger and health risks in disaster sites, this team has taken the first steps towards developing a system to make legged robots walk swiftly and agilely, applicable on the typical rough terrain in disaster sites. This is possible because this project has enabled Cassie to transition from walking to stepping onto a static platform — the foundation needed for Cassie to step onto the dynamic platform, the Hovershoe. By implementing a two-camera system and enhancing gait synthesis, Cassie has more fine-tuned control over its movements to better assist first responders.
We had a chance to speak with the capstone team members Jamie Chen (ME), Sonny Li (ME), David Tondreau (ME), and Mengyue Wang (ME), about their experience.
How did you define the scope of the Capstone project?
“In many ways, the scope of our project was set by the 2019 MEng team who successfully enabled Cassie to ride Hovershoes. Currently, there are separate controllers that enable Cassie to robustly walk and to ride Hovershoes, but no controller to bridge these existing works. Thus, our scope was originally to connect these two functional controllers to enable Cassie to autonomously walk up to a pair of Hovershoes, step on, ride, and then step off at the destination.
However, once we realized how difficult this goal was to accomplish within our time constraint, we utilized an interactive approach for scope definition by continually discussing with our advisors the challenges we faced and restructuring the scope of our project iteratively with their input. Eventually, we revised the final milestone to be Cassie stepping onto a static platform, as we found the technical realization of such to be both a novel contribution and a substantive step toward the goal of stepping into Hovershoes.
How exactly can your robot aid first responders in disaster cases? What functions can it serve?
Although modern technology has better equipped first responders, humans still must physically enter dangerous disaster and emergency areas. There are numerous examples: the Sago Mine disaster, the Fukushima nuclear meltdown, the California Wildfires, the search for survivors after 9/11, the 2010 Haitian earthquake, etc. Two commonalities in all of these examples are: first, a need to respond quickly to mitigate damage and save human lives, and second, a complex terrain inherent at the disaster site.
The Cassie robot is uniquely positioned to address both of these issues. In our project, we developed a controller to enable Cassie to step onto a static platform, laying the foundation for Cassie to transition from walking to riding Hovershoes autonomously. Once enabled, Cassie is fast enough to respond to a disaster scenario and mobile enough to navigate over complex terrain.
“In our project, we developed a controller to enable Cassie to step onto a static platform, laying the foundation for Cassie to transition from walking to riding Hovershoes autonomously.”
Once bipedal robots are fast and mobile enough to respond to disaster and emergency scenarios, a host of sensors and tools could be incorporated for the specific scenario. For example, in the Sago Mine disaster, first responders were delayed by twelve hours due to high levels of carbon monoxide and methane gas. Most topically, with the current global coronavirus pandemic, first responders risk contracting the virus. In all of these examples, Cassie could be outfitted with the appropriate tools and sensors, urgently respond, walk over the complex terrain, and mitigate the situation while limiting the risk to first responders.
Are there other robots similar to yours? If so, what makes your robot distinct?
Many have likely seen the mind blowing work done by Boston Dynamics, specifically the Atlas robot, or have heard of the Defense Advanced Research Projects Agency (DARPA) Robotics Challenge and think, “Oh, these are all the same because they are bipedal robots, yours just doesn’t have a torso!”
However, a key difference is that those robots have motors at the ankles to maintain stability. Cassie, on the other hand, has neither motors at the ankle nor a square foot platform to control lateral stability. Instead, Cassie is unique because its lined feet enable better stability in truly rugged terrain, commonly found in disaster sites.
While legs provide adaptability, wheels provide speed. Some robots such as ANYmal and DRC-HUBO have been mechanically designed with both legged and wheeled ligaments to switch between, however these robots are not optimized for either locomotion.
What was your most significant teaming challenge & how did you tackle it?
In Fall semester, we divided our team into subteams in which every individual member focused on tackling a specific problem. Everyone worked independently most of the time and met up with our graduate student mentors once a week for updates and discussions. During these times, we had little knowledge what other team members were working on.
After reviewing the team reflection for ENGIN W270K at the beginning of Spring semester, we all agreed that we needed to communicate and collaborate more. In order to work as a whole team, we decided to set up “Lab Hours”, when we would work in the lab together four times a week. The frequent lab hours were a significant change to our team and brought us together! We were able to share our insights and thoughts instantly, and our fragmented work felt more like one project again.
What was your most significant project management challenge & how did you tackle it?
With about three months left before graduating, our technical progress was not showing the results we wanted. Our team had heard one of our Hybrid Robotics Lab mentor was utilizing a different approach to low-level control of bipedal robots, Contact Force Optimization, as opposed to the PD-based control approach we had been utilizing all year.
Our team was divided — while we thought we may have been able to progress faster using the Contact Force Optimization approach, we knew it meant learning and implementing a new approach with little time remaining. After inviting everyone to get together on a call — our team, graduate mentors, and key stakeholder Professor Sreenath, we all discussed what we had gathered and to ensure everyone was on the same page about the roadblocks we were encountering. Unanimously, our team voted to follow the advice of our key stakeholder and graduate mentors and to stick to the PD-based control approach we had been utilizing all year. Moreover, soon after this decision, we made a few technical breakthroughs which greatly sped up our progress.
What was your most significant technical challenge & how did you tackle it?
Thanks to science fiction movies like iRobot or Terminator, it’s so easy to imagine a world full of humanoid robots. Moreover, I love looking at old futuristic photos from the 40s to 60s era (think The Jetsons), where robots and flying cars are rampant. However, while we do have robots able to perform gymnastic routines thanks to Boston Dynamics, if you take a look at the DARPA robotics challenge, you can see our best robots currently still struggle in unstructured environments.
The ugly truth is that legged robots, especially bipedal robots, are really hard. That said, it is that challenge which excited and motivated our team. Thus, we had many technical challenges especially as none of our team entered the project with prior experience in the field of legged robot control.
Our biggest hurdle was implementing acyclic double support phases into the gait optimization program and low level controller. Given issues like model mismatches, finding the right constraints and associated control parameters proved to be a challenge. Nevertheless, our team was able to develop a solution to this challenge by iteratively working together and sharing knowledge.
Have you had a capstone project in your undergrad & if so, how has this capstone experience been different?
We all had some sort of senior-capstone project during our undergrad. What sets the MEng capstone experience apart is our ability to collaborate and seek mentorship from true experts in our capstone’s technical field.
“What sets the MEng capstone experience apart is our ability to collaborate and seek mentorship from true experts in our capstone’s technical field.”
In the end, legged robots are still a pretty small subset under the robotics industry, and to be able to shoot an instant message to the senior Ph.D. members of the lab, then get expert advice from the world’s leading researchers was simply a fantastic learning experience.
You probably had a number of course-based projects during your MEng year. How is the capstone different?
I think our capstone is a bit special because as much as it is an engineering project, it is also a research project. After speaking with some of the Ph.D. students in our lab, we found out that the problem we are trying to tackle is novel in the sense that it has not been done in academic literature. For this exact reason, the capstone project is a much more open-ended engineering problem. Our advisor gave us general guidance but the technical decisions and ultimately, the deliverables were up to us to create.
“I think our capstone is a bit special because as much as it is an engineering project, it is also a research project.”
On the other hand, the course-based projects are much more closed-ended problems. For example, we had projects such as: utilizing MPC to solve an engineering challenge, designing a controller with cascaded control to enable a drone to land on a target, and using an estimation algorithm to estimate the pose of a vehicle. While all these projects are somewhat ambiguous, there was a structure we could systematically follow to complete the project. Thus, our Capstone project posed a more realistic problem, similar to what we would face in the real-world.
Is there anything else you’d like to add?
Shout out to the Hybrid Robotics Lab! We expect innovative things from the bright researchers there! Our team is especially grateful for Professor Sreenath and Bike Zhang’s leadership and mentorship. Thank you!”