Cal Poly Pomona Autonomous Touring Experience Prototypes Driverless Vehicles




​Graduate student Xianmei Lei watches the yellow four-wheeled vehicle make its way along the walkway outside Building 8. It looks like a robot and runs on a robot operating system but it’s more than that. It’s a prototype for an autonomous vehicle. CATE—Cal Poly Pomona Autonomous Touring Experience—is different from other driverless vehicles in that the finished product will be an electric cart designed to drive on college campus walkways. The project is a collaboration of Pomona’s computer science department in the College of Science, the College of Engineering and the College of Business. 

Engineering is providing the low-level controls on the project. “The low-level controls activate all the systems such as motor, steering and braking. There are sensors that acquire data about what’s happening. That data is processed and then sent to the high-level controls, which is what computer science is handling,” Dr. Scott Boskovich, assistant professor of engineering, explained. 

Lei, a graduate student in computer science, is the software lead. “In this project,” she said, “the software we’re writing deals with four areas: First there is localization, which tells the vehicle where it is. Second, there’s mapping, which tells it where it’s going. Third is low-level control that tells it how to get there. And fourth is exploration, which the vehicle needs in order to understand its environment.” 

The prototype vehicle is built on an unmanned ground vehicle (UGV) called a Husky. It uses multiple sensors that include two cameras, a GPS, an inertial measurement unit (IMU) and a light detection and ranging (lidar) sensor. 

“This research is very different from other programming in that the way to get there isn’t clear. It requires a lot of research,” Lei said. One of the biggest challenges is compensating for noise. “Noise is the error that’s introduced into the system by the sensors,” she explained. “These errors can add up very fast, creating major problems. For example, with the encoder that counts wheel rotations, a slight error will keep compounding and throw off localization so that the vehicle won’t have an accurate idea of where it is.” 

“GPS can be affected by atmospheric conditions but even without that interference, GPS has a 10-meter error range, which is too large a margin of error.” 

Once all sensor noise or errors are considered, that data then needs to be combined to guide the vehicle. The team uses an unscented Kalman filter, an algorithm to combine the data from multiple sensors to gain a more accurate picture of where the vehicle is. 

Donors help make research like this possible. In this case, the donor, who also funded several other projects, wishes to remain anonymous. The College of Science has an initiative called Discovery Through Research that raises money to provide more research opportunities for students. With increased funding, the College of Science is able to offer more research projects to students. The opportunity for students to conduct research, acquire and analyze data and communicate their findings can be transformational in their development as scientists. 

Dr. Amar Raheja, computer science professor, said: “As faculty adviser to Lei, I help in assessing her research goals and I’m there as a resource, but this is very much a student-led, student-driven project that prepares students to conduct research and solve problems just like they will need to do when they leave here and are working in the field.” 

Professor and Computer Science Department Chair Daisy Tang said: “The prototype is built on the Husky platform but the final product will be an electric cart. We now have the cart and we’ve started purchasing the sensors for it, which we expect to install next year.” 

In addition to leading the software portion of the CATE project, Lei is interning at the Jet Propulsion Laboratory. She’s working on a Defense Advanced Research Projects Agency (DARPA) Subterranean Challenge. Her work focuses on object detection for underground exploration and rescue. “My work uses neural network or deep learning with thermal and clothes detection data to aid in identifying survivors,” Lei said. 

The research will make it easier for rescue crews to find survivors in cave or mine disasters