Baxter the robot provides unique educational, research opportunity

Baxter, the approximately 6-foot, 300-pound, two-armed robot that lives in the basement of CSL, has been the focus of research for multiple graduate students since taking up residence in CSL about two years ago.

Aerospace Engineering Associate Professor Tim Bretl and his students have been using Baxter, an economical robot designed for small business manufacturing, to develop algorithms for robotic manipulation and perception of flexible objects. The students, many from different engineering fields, have worked with Baxter on a variety of projects, from writing code that allows him to complete challenging tasks to using him as a model for prosthetic limbs.

“Robotics is inherently an interdisciplinary endeavor,” said Kevin Chen, a master’s student in electrical and computer engineering who works with Baxter. “There are many different fields that go into understanding robots, so having a multidisciplinary team is definitely helpful in designing and working with robotics.”

For example, while Aerospace Engineering Ph.D. student Andy Borum works on mathematical theories that will allow Baxter to manipulate deformable wires and cables, Chen and Jacob Wagner, a Ph.D. student in mechanical engineering, are working on improving Baxter’s accuracy by identifying sources of error within Baxter.

“Picking and placing objects that are rigid is a classic problem in robotics, but manipulating flexible objects, such as wires and cables, is a much more difficult task,” Borum said. “The long term goal is for Baxter to manipulate those flexible objects.”

Baxter’s original design is not as accurate as more expensive robots, but rather, he is designed for pick-and-place type operations in manufacturing, while also being adaptable to learn new movements and objects. In addition to being less expensive, he also has the ability to interact safely with people, due to features such as compliant joints and sonar sensors that can detect nearby objects. Due to these unique features, he is an ideal robot for experimentation and learning.

“Baxter was built to be compliant and to be around people,” Wagner said. “He can detect when he has hit something and he can stop moving when he detects contact, which is not true of some industrial robots.”

The lab purchased Baxter during the summer of 2013, but since they began last fall, Wagner and Chen have been the first students to be fully devoted to his development.

“One of the things I’m interested in is for Baxter to be an assistant to a human,” Wagner said. “I don’t know in exactly what capacity yet, but possibly using his robotic hands to present and pass objects to humans.”

Wagner would like to use machine learning techniques to allow Baxter to complete tasks such as handing objects to humans and learning how to relate Baxter’s movements to his sensory observations obtained from the cameras and sensors installed in him.

“We want to analyze the relationship between the movement commands we send to Baxter and the information we obtain from his sensors,” Wagner said.

Other students have used Baxter for specific projects, such as computer science Ph.D. student Jessica Mullins, who taught Baxter how to sort colored balls into colored buckets.

Mullins wrote code that allowed Baxter to detect objects and colors using the camera in his hands and he was able to sort the balls into their coordinating colored buckets.

“In computer science, we’re not working directly with robots, but I thought it was interesting and thought I’d take advantage of people in my lab who know about them,” said Mullins, who is also helping Wagner and Chen on the precision research. “My area of expertise is more human-computer interaction, so that’s one of the reasons we’re working on the human-robot type interaction task.”

Aadeel Akhtar, a Ph.D./MD student in neuroscience, also worked with Baxter to map neural signals coming from a person’s muscles to Baxter’s joints.

As Akhtar’s research focuses on neural interfaces and controlling robots with neural inputs, he used Baxter as a proxy for a prosthetic device. Based off of those studies, they were able to design their own prosthetic devices that are mobile and have sensory feedback and muscle control.

“Baxter is really a great platform to work with because a lot of the software he uses is standard, so it’s very easy for us to interface him with the software we had written for doing neural control,” Akhtar said. “Other robots we worked with have been harder to interface with.”

While all the students may be using Baxter for a variety of purposes, any improvements made will benefit all the students working with him. In addition to collaborating together with their research, the students are excited that their work could have applications in many outside areas, as the precision and tracking work they are doing can be applied to other robots.

“We’re using Baxter as a platform to test general robotic manipulation algorithms, but our results could be used on any robot, not just low-cost manufacturing ones,” Wagner said.