February 24, 2016      

Even though naval vessels are surrounded by water, a fire aboard a ship is one of the biggest dangers to sailors. Fuel, ordnance, and tight quarters make firefighting extremely difficult.

The U.S. Navy has awarded $587,115 to Dmitry Berenson, an assistant professor of computer science and robotics engineering at Worcester Polytechnic Institute (WPI), to develop software for a firefighting robot. The three-year grant is an extension of research begun at Carnegie Mellon University (CMU) and the University of Pennsylvania.

Virginia Polytechnic Institute and State University (Virginia Tech) has been working on the hardware for SAFFiR — short for Shipboard Autonomous Fire-Fighting Robot — a humanoid designed to operate in spaces designed for human sailors.

“We are quite excited to be part of this,” Berenson said. “Only a few schools were considered. The Navy wound down the first stage of its work: building the robot, conducting a preliminary demonstration of spraying fire with a hose, and walking a little.”

Achieving autonomy

“We’re looking to develop the technology a little further,” he explained to Robotics Business Review. “We stood out because of our focus on autonomous motion-planning algorithms. Instead of active teleoperation, we’d like humans to do as little as possible, directing the robot rather than steering every move.”

“For instance, if I say to the robot, ‘Walk from the mess hall to the engine room on the ship,’ it should just be able to do that,” Berenson said. “There should be no need for instructions on how to go up stairs or through a door.”

Although the robot is at Virginia Tech, WPI’s team is testing its route-planning model in simulation in Massachusetts before continuing on a decommissioned naval vessel. The team includes undergraduate student Will Pryor, graduate student Yu-Chi Lin, and postdoctoral fellow Andreas Orthey.

Virginia Tech conducted a demonstration of SAFFiR in 2014 aboard the U.S.S. Shadwell, and WPI has helped make progress since then. “We’ve been working for a little more than half a year on this project,” said Berenson. “We’ve been able to several key things, laying a foundation for things to come.”

The SAFFiR researchers have improved the robot’s interfaces, enabling the planning software to communicate with onboard components. They have also been working on perception, using a laser scanner to provide information on where the robot can place its hands and feet.

“Brian Lattimer, [an assistant professor of mechanical engineering] at Virginia Tech, has been working on sensing that works well when smoke occludes vision,” Berenson said. “For our purposes, it’s about using a laser scanner/lidar; with a stereo pair, we have what we need for navigation.”

In addition, Berenson and his collaborators have created methods to quickly plan footstep sequences with different-sized obstacles, and they have been investigating the fundamentals of how the humanoid robot can respond to forces coming from different trajectories, such as waves rocking a vessel or falling debris.

“There are two challenges of autonomous navigation that we’re focusing on now,” he said. “It’s in a complex, cluttered, changing environment, and the ship is rocking. We treat it as a set of possible disturbances — this set is what the robot needs to think about when it’s planning its motion, what magnitude of force it needs to resist. We’re working on having it plant its feet and use its hands for steadying as a human would.”

The Office of Naval Research is sponsoring other studies on SAFFiR’s sensors, fire-suppression aspects, and other maintenance tasks it could perform.

“Rescuing people is a great application, but we’re not focused on that right now,” Berenson said. “We need more development of moving and gripping.”

Why a humanoid?

Berenson acknowledged that a humanoid design is challenging and that wheeled robots are generally more energy-efficient or stable. “It’s true that, at last year’s DARPA Robotics Challenge, NimbRo had six wheels, but there were some elements of the contest that were not realistic, like rough terrain and walking,” he said.

“[NimbRo] drove straight through the clutter, but if I rearranged that clutter differently, that robot wouldn’t be able to traverse it,” Berenson observed. “I don’t think we want to overgeneralize the results. A conventional mobile robot would be too wide or couldn’t widen its base of support easily.”

“There’s a critical need for a robot with arms and legs … walking through very cluttered environments, with height changes, steps, and ‘knee-knockers,’ or thresholds,” he said. “In a submarine, we’d like it to be able to climb small ladders and move through really constrained environments designed for humans.”

WPI and CMU had a joint entry at the robotics challenge, placing in the top third.

Virginia Tech had an entry at the robotics challenge, ?but it didn?t do that well,? Berenson recalled. ?It did demonstrate an important capability — it walked the whole distance on sand rather than driving in a car. Not a lot of robots can do that.?

“Our robot is built to be compliant, with elastic actuators, which is really important in a shipboard setting,” he said. “It can absorb changes in topography easily, mechanically responding naturally to the surface.”

“By contrast, [Honda‘s] Asimo and DRC winner HUBO are precise but not very compliant. If unexpected things occur, it’s harder to compensate, and they need need active sensing,” Berenson explained.

First, faster steps

Berenson noted that he got involved with humanoid robots during his studies with James Kuffner, who spoke about autonomous route-planning at the Robotics Conference at this year’s Consumer Electronics Show. Kuffner was a professor at CMU and director of robotics engineering at Google Inc. He is now leading artificial intelligence research at the new Toyota Research Institute.

Although most current androids move more slowly than humans, researchers are working on making them run, Berenson said. “We need to keep making advances in that direction; we’re interested in using the capabilities in the planning algorithm.”

“My lab is working on general problems — we don’t want to just make a special-purpose firefighting robot,” he added. “These algorithms can be used in a wide range of scenarios. Maybe the robot is in a cluttered factory, conducting disaster response, or in a furniture store. We’re not restricting it to a particular size corridor or door.”

The National Aeronautics and Space Administration is offering $10,000 for the best algorithms in its Robot Vision Tool Manipulation contest. NASA’s Robonaut 2 is currently being tested on the International Space Station (ISS).

“Space is a great application, and a robot would need to make contact with its hands to move,” Berenson said. “We’re definitely planning for our robot to ascend and descend stairs. Ladders are further out, because the arms must be able to bear the weight of the robot.”

Speaking of weight, SAFFiR has some onboard computing, “but at this stage of development, we’re working on just getting the algorithms to work correctly,” Berenson said. “We’re planning on a desktop sending commands.”

“The robot will work with humans and must be able to navigate by itself,” Berenson said. “With the current training of Navy firefighters, it’s a team, with one person guiding others. My understanding is that the robot would go ahead where it’s dangerous.”

More humanoids step up

NASA recently loaned two of its R5, or “Valkyrie” robots, to MIT and Northeastern University for development. The agency also awarded the Boston-area universities $250,000 to develop algorithms for autonomy in dynamic environments.

“NASA is counting on robots to setup and care for deep-space exploration facilities and equipment pre-deployed ahead of astronauts,” said Sasha Congiu Ellis of NASA’s Langley Research Center. “Robots are also excellent precursors for conducting science missions ahead of human exploration.”

As part of the renewed rivalry between the U.S. and Russia, the Russian Foundation for Advanced Research Projects this month showed off two robots also designed to work on the ISS.

Meanwhile, Japan’s National Institute of Advanced Industrial Science and Technology and France’s Multi-Contact Collaborative Humans in Aircraft Manufacturing have partnered with Airbus Group SE on developing humanoid robots for aircraft manufacturing.

The four-year COMANOID research project, which is partly sponsored by the European Commission’s Horizon 2020 program, is also working to improve route planning, manipulation, and autonomy in confined spaces.

In addition, Boston Dynamics, a unit of Google’s Alphabet Inc., this week demonstrated its latest iteration of its Atlas robot. A previous design was among the robots mocked at last year’s DARPA Robotics Challenge for being unsteady.

However, this version is lighter, more agile, and faster than its predecessors. While the latest Atlas doesn’t have a tether, it’s unclear what level of autonomy it has.