After a decade of work, the Robotics Collaborative Technology Alliance marked a milestone in its autonomous mobility and robotics research for the U.S. Army this month at an event at the National Research Engineering Center in Pittsburgh.
The Robotics Collaborative Technology Alliance (RCTA) has brought together government, academia, and industry to help the Army attain its research and development goals around unmanned ground vehicles. The consortium started with $63.2 million and a five-year plan, and its corporate members included General Dynamics Land Systems, Boston Dynamics, and QinetiQ North America.
The alliance’s academic members included Carnegie Mellon University, the Jet Propulsion Lab/California Institute of Technology, the Massachusetts Institute of Technology, the University of Pennsylvania, the University of Central Florida, and Florida State University.
Robotics research addressed ‘fundamental problems’
Founded in April 2009, the robotics research organization initially identified four areas as critical to autonomous military systems: perception, intelligence, human-robot interaction, and dextrous manipulation and unique mobility. This includes a focus on basic research around mobility.
“We looked at fundamental problems for meeting multi-domain operations, competing against a nuclear adversary in air, on land, at sea, in space, and in cyber,” said Dr. Stuart Young, collaborative alliance manager at the RCTA. He is also program manager for artificial intelligence for maneuver and mobility at the Combat Capabilities Development Command in the Army Research Laboratory (ARL).
“Specifically, we were looking at autonomy on the ground with less teleoperation. Off-road is a very complex environment in a disaster or war-torn situation,” Young told The Robot Report. “Soldiers need teammates that operate robustly and have a common understanding of the mission.”
Differences from self-driving cars
The RCTA’s robotics research included a close relationship with other efforts. “We’ve worked with DARPA on supporting and evaluating challenges,” said Young. “A lot of the teams are the same performers we’ve worked with on a foundational level.”
“Clearly, the private sector will outspend us 10 to 1, but we’re trying to solve different problems,” he said. “We don’t develop our own hardware chips or sensors, and we spend most of our efforts on developing unique algorithms. Wherever possible, the ARO [Army Research Office] works with industry partners.”
“While there have been huge advances in neural networks and perception, there are still barriers to making these things fieldable with soldiers,” Young said. “Autonomous car companies are using techniques we developed, but they require a massive amount of pre-mapped, a priori information. Machine learning needs a lot of data to perceive.”
“As these companies train cars to operate in one type of environment, such as Pittsburgh or San Francisco, they can’t just go somewhere else with those algorithms,” he added. “There’s always another edge case.”
“We’ve had to develop new techniques for operating in unstructured environments without a priori data,” Young explained. “Autonomous vehicles have to plan for 3D space — if it comes to water, it has to reason about navigability. Cars don’t have to wonder whether a road is connected or there’s a ditch.”
“Cars just have to get from Point A to Point B; they don’t have to perform search-and-rescue functions,” said Young. “We need more of a shared understanding of what your partner is doing, like two firefighters. Robots and people think differently about the world, so we’ve done a lot of work in natural language processing.”
“It’s very exciting — we’ve tried a holistic cognitive architecture to understand concepts,” he said. “We’ve developed an architecture to not only parse language, but also ground it in common symbols so that people and robots can perform tasks together.”
For instance, a robot might recognize an object as a fruit stand, but a human must be able to tell it to meet there. “The robot might recognize the 10,000th rock or coordinates that wouldn’t make sense to the human, but the robot could know the concept of a fruit stand and then the concept of fruit,” Young said.
“To sense the difference between two barrels, one of which is full, and the other of which is empty, the robot must manipulate the environment to understand,” he explained. “It may have to take action to disambiguate, and the architecture can absorb and reason.”
The Distributed and Collaborative Intelligent Systems and Technology Collaborative Research Alliance (DSIST CRA) is exploring “heterogenous teaming,” in which a mix of robots, drones, and autonomous vehicles can collaborate.
“It’s like a soccer team — players must share information and have common objectives,” Young said. “It complements the RCTA, but the problems are different. DSIST’s work will inform the RCTA’s work on a single platform.”
Mobility and manipulation
While humanoid robots still draw a lot of attention, much of the robotics research the lab has done has been “platform-agnostic,” Young said. The RoMan or Robotic Manipulator has been tested for debris removal.
In collaboration with NASA’s Jet Propulsion Laboratory and the Institute for Human and Machine Cognition (IHMC), the RCTA is developing the Legged Locomotion and Movement Adaptation robot, or LLAMA.
“We’ve been exploring a quadruped that operates with a human. We need a platform that can go where they can go. Legs give the ability to operate over discontinuous surfaces, like leaping over a ditch or stepping on stones in a pond,” noted Young. “A dismounted soldier needs confidence that a robot carrying food or ammunition can keep up when a wheeled or tracked vehicle can’t. Survivability is paramount.”
“The Army has asked us to scale up designs, so maybe you’d have autonomous tracked and wheeled vehicles without people in them as well as robots that could climb,” he said.
“We’re working on mobile manipulation of unknown objects in the wild,” Young said. “We’re trying to generalize to more complex environments more robustly. We’re not just extending brute-force algorithms and are looking for more efficiencies.”
The Robot Report is launching the Healthcare Robotics Engineering Forum, which will be on Dec. 9-10 in Santa Clara, Calif. The conference and expo will focus on improving the design, development, and manufacture of next-generation healthcare robots. Learn more about the Healthcare Robotics Engineering Forum, and registration is now open.
Robotics research leads to a general-purpose robot?
Many experts in robotics research agree that artificial general intelligence or a general-purpose robot is a long ways off. “It depends on how hard the task is,” said Young. “If you can define it well, we’ve gotten pretty good, but we need it to be robust.”
“But things can have different meanings depending on the context,” he added. “The capability of putting things in context is quite challenging, and AI systems are brittle. As we progress toward generalized capabilities, we could spin off robots with more limited capabilities, like self-driving cars.”
“Systems that are deployable in constrained environments are pretty good, but home robots have lots of complexity and objects to deal with,” Young said. “On the Army side, we need systems to be robust and resilient. If Netflix recommends a bad video, the ramifications are much lower than making decisions of where to have a soldier on a battlefield or to look for a child in a collapsed structure after a hurricane. We need a higher level of confidence.”
Marching to the next mission
After 10 years, the RCTA may cease in name, but robotics research will continue. “ARO research is more basic than applied, and the CTA was trying to do things very rapidly,” Young said, adding that projects are already in the pipeline to partners in Army centers and shops.
“The RCTA has been rather broad in scope, and it has lasted 10 years,” he observed. “The Army Futures Command has identified its priorities for the next-gen platform and the specific capabilities it’s interested in.”
“We’re looking for off-road autonomy for the next generation of armored combat vehicles,” said Young. “Instead of 10,000 problems, we’re looking at the 100 that the team is interested in. That partnership makes for more efficient use of taxpayer dollars. We’ll be able to do things in a shorter time frame and evaluate progress made, and we’ll be announcing some new initiatives soon.”