Topcon Positioning Systems, Inc.

The Topcon team with a small-scale bulldozer prototype used in their project. Left to right: Kiefer Selmon, James Trombadore, David Kooi, and Donald Avansino. (Photo by Tyler Bartholome)

 

The Topcon team with a small-scale bulldozer prototype used in their project. Left to right: Kiefer Selmon, James Trombadore, David Kooi, and Donald Avansino.

(Photo by Tyler Bartholome)

 

UC Santa Cruz students use advanced computer vision techniques to enable autonomous approach-and-dig operation for Topcon bulldozer

 

June 20, 2018

By Erin Foley, Tyler Bartholome, and Karyn Skemp

 

The Baskin School of Engineering’s Corporate Sponsored Senior Project Program (CSSPP) provides students with a unique opportunity to experience working on real-world engineering projects as part of their undergraduate education. Throughout the academic year, students interact with teammates, sponsors and faculty. Some make visits to their corporate sponsor’s worksite, and all are required to solve problems along the way. By working with mentors at corporate partner companies, students learn important skills, take on interesting challenges, and begin to understand what it means to be a professional engineer.

 

One of this year’s projects was sponsored by Topcon Positioning Group, a provider of high-precision GNSS positioning technology for construction, geopositioning and precision agriculture. Among Topcon’s offerings are several applications that support automation in construction vehicles. Students James Trombadore, David Kooi, Kiefer Selmon, and Donald Avansino worked with Topcon to research and develop technology that would enable a bulldozer to autonomously identify, approach and dig a stockpile at a construction site.

 

Topcon provided the team with a small-scale bulldozer prototype called the HouseCat mkII. The students had just four months to create the technology that would give this prototype its autonomous functionality.

 

The team broke the problem down into three distinct components: identifying and aligning with a stockpile; determining the optimal approach to the stockpile; and performing and verifying the dig. They mapped out a solution that relied on sophisticated concepts from computer science and engineering, including machine learning, computer vision and neural networks.

 

 The Topcon team field testing their autonomous bulldozer technology.

 

The Topcon team field testing their autonomous bulldozer technology.

 

To accomplish the first task -- getting the application to identify and align with a stockpile -- the team created a neural network, which they trained using hundreds of images of stockpiles of variable size, shape and material. They installed three cameras onto the HouseCat: a 2D camera for object detection and two additional “stereo cameras”­­ for depth perception. With these sensory inputs, the application can generate a spatial map of the area in front of the bulldozer. The map, in conjunction with an optimal approach algorithm written by the students, tells the bulldozer where to enter the stockpile for the best dig. Linear actuators on the arms of the bucket enable the mechanical dig action. The filled bucket is then tilted up in front of the center camera and the image is processed analyzed. If the bucket is at least 60% full, it meets the minimum threshold requirement, and the bulldozer traverses to a dump site and returns for another dig. If the minimum threshold is not met, the bulldozer realigns and digs again.

 

The team attributes it success to hard work, a willingness to take chances, guidance from faculty mentor David Munday, and insight from the Topcon engineers.

 

Munday, a key member of the CSSPP faculty team, was happy with the team’s results. “The key innovation our students produced this year is the creative use of cutting edge computer vision techniques to overcome sensory limitation,” he said. “Unlike standard autonomous vehicle sensing systems, this system does not have expensive proximity sensors, LIDAR, or even weight sensors to aid in bucket load detection. Like a human being that loses one of their senses, the students’ vision algorithms became sharper to adapt beyond those limitations.”

 

Donna Kelley, senior product design engineer at Topcon said working with the students was inspirational. “The engineers who worked with the students were energized. And the students learned a lot about real-world integration and about how engineering works in the real world.”

 

“I wouldn’t hesitate to do this again,” Kelley said. “It was a great experience. We all got a lot of out of it.”

 

About Topcon Positioning Group 
Topcon Positioning Group is headquartered in Livermore, California, U.S. (topconpositioning.com). Topcon Positioning Group designs, manufactures and distributes precision measurement and workflow solutions for the global construction, geospatial and agriculture markets.