Researchers Tap Robots for the Dirty, Dull and Dangerous

Although the human directing the robot provides the high-level instructions — perhaps by clicking on the object a robot is supposed to engage with — it’s the robot’s job to translate those commands into action and to move itself to perform the job.

“It turns into a planning and control problem where the robot needs to take into account its understanding of the 3D environment, its safety and the human’s wants and then synthesize that into a plan,” said Atanasov. “It’s a major task; we’re interested in further developing synthesized robotics that streamline the process and make human-robot interactions smoother.”

Visualizing future use cases

Still, for robots, this type of visualization is a relatively new way of thinking.

Historically, robots would simulate the cause and effect of each of their potential actions to choose the best option for their task and to avoid collisions, explained postdoctoral researcher Brian Lee, who is part of Atanasov’s lab, which is affiliated with the UC San Diego Contextual Robotics Institute. Now, using visual mapping and 3D reconstruction, robots can identify everywhere they can safely move or engage with an object.

Coupled with human monitors who can identify potential dangers and correct a robot’s actions, that ability to move safely in an unstructured environment is exciting, Lee said, adding that it “opens up a greater range of possibilities and robotics applications.”

From a construction standpoint, this flexibility and awareness can lend itself to robots playing a bigger role in bringing projects from idea to reality — especially as modular design grows. This type of building slots together from prefabricated components like Lego blocks can produce a variety of projects faster and at a cheaper price point.

“I see robots playing a bigger role in this type of construction because many of those components will be made in advance,” Atanasov said. In his eyes, that will make it easier for robots to build and execute projects. “The humans can direct the operations and supervise the overall process. Or, we could have a scenario based on human-robot collaboration, where the robot does the brunt of the work and the human manages the more fine-grained manipulations.”

More broadly, however, the mapping and processing technology the researchers are developing can also be used to make high-quality virtual reality more achievable. Currently, VR requires sophisticated algorithms and extensive filming. But, with highly detailed scanning and cameras at certain positions, Nguyen explained, a few strategic snapshots can quickly and accurately reconstruct a scene.

“Real estate agents want something like this that could let them scan a house in real-time and get 3D models of the rooms,” he said. “They could then show these reconstructions to potential buyers or renters who could explore an apartment’s layout virtually without ever having to visit.”

And, he’s confident that their technology can be that solution.

“We have the fastest, most accurate options for 3D reconstructions,” Nguyen said, “and we are very competitive in this area.”

What’s next

Another year and a half remain of the grant funding. What’s next for Nguyen, Atanasov and the rest of their robotics research crew?

Integration.

“We’re very happy with the results for the 3D construction, but we need to integrate that technology with robotic manipulators,” Lee said. “Now, we need to test it with a human operator and integrate our system with a compelling setup that can interface with a human.”

Although their technology will first be tested by funder ITOne in an apartment construction project using robotic arms, the team sees broader applications for the work beyond the construction industry.

“A lot of the research and algorithmic formulations are more general than this problem,” Atanasov said. “A robot being able to plan its motion and understand its surroundings can be applied to scenarios ranging from hospitals and warehouses to autonomous vehicles and agriculture. Even though they seem like completely different domains, the types of robotic processes are similar.”