May 11, 2005
NASA Developing Space Robots with Human Traits
NASA -- Will robots that help astronauts in space be as friendly and likeable as the fictional "R2-D2" android portrayed in the original "Stars Wars" motion picture?
NASA scientists say robots will behave more like human beings in the future, even if - like R2-D2 - these machines do not look like people.
NASA researchers envision futuristic robots that 'act' like people, enabling these mechanical helpers to work more efficiently with astronauts. Human-robot cooperation, in turn, will enable exploration of the moon and Mars, and even large-scale construction in extraterrestrial places, according to scientists. Because human crews will be limited to small teams, astronauts will need robot 'helpers' to do much of each team's work, scientists reason.
"Our goal is not for robots to have the same 'thought process' as humans, but rather for them to act, respond and interact more 'naturally' in ways that humans do with other humans. This requires that robots possess traits such as self-awareness (recognition of their limits and when they need to ask for help), and human-awareness (knowing to whom they are talking, and when it is an appropriate time to ask a question)," said Illah Nourbakhsh, a scientist who leads a group developing human-robot teams at NASA Ames Research Center in California's Silicon Valley.
Though remotely-controlled machines and robots that work entirely on their own are valid goals, the NASA-Ames team plans to focus on robots that are partly controlled by people and operate independently the rest of the time, Nourbakhsh said.
There are three main areas that Nourbakhsh's team will develop. One is called 'collaborative control,' during which the human being and the robot will speak to one another and work as partners. "A key benefit of collaborative control is that the robot is able to ask questions of the human in order to compensate for (the robot's) limitations," Nourbakhsh explained.
A second area that NASA Ames researchers will develop is to build robots that have reasoning mechanisms that work similarly to human reasoning. "Of primary interest is making the human and robot understandable to each other," Nourbakhsh said.
Thirdly, the researchers will conduct field tests of people and robots working together. Many experiments also will occur in a special, indoor laboratory under construction at Ames, featuring a control room with a window looking out on robots working in a large area that will simulate the surface of a moon or planet. The control room will imitate a human habitat on the moon or Mars. Scientists are targeting completion of the robot laboratory for the fall of 2005.
The robots will help assemble buildings, test equipment, weld structures and dig with small tools. Human-robot teams will use a checklist and a plan to guide their joint efforts, according to NASA Ames scientist Terrence Fong. Human beings and robots will use "partner-to-partner" interaction to share information and to support one another. "It is the way human construction and maintenance crews frequently operate," Fong explained.
"We will perform detailed workflow and critical incident analysis to understand the impact of partner-to-partner, human-robot interaction on tasks, to identify failure modes and learn how to improve execution," said Nourbakhsh. "We will place significant emphasis on assessing system performance, human performance and robot performance," he added.
The group's robot development work will focus on specific tasks essential for basic exploration mission operations including: shelter and work hangar construction, piping assembly and inspection, pressure vessel construction, habitat inspection and resource collection and transport. "Each of these operational tasks demands effective human-robot teamwork and requires extensive interaction between human and robot," Fong observed.
Scientists say human-robot cooperation will result in a better outcome than human- or robot-only teams could accomplish.
To make human-machine teaming a reality, a NASA multi-pronged effort is underway to develop robot intelligence. Similar to human thinking, it's designed to improve the mechanical workings of robots and to standardize human-robot communications.
"We believe that by building robots with reasoning mechanisms and representations that are similar to what humans use, we can make human-robot interaction more natural and human-like," said Nourbakhsh.
"These teams will include humans and robots working in close quarters as well as humans controlling and interacting with robots over medium distances and from ground control," Nourbakhsh predicted. "The effectiveness of human-robot interaction will have a major impact on productivity, cost and risk reduction in exploration missions," he explained.
"Conventional human-robot dialogue is limited to 'master-slave' commanding and monitoring," Fong said, speaking of how robots generally have been controlled to date. "As a result, system performance is strictly bound to the operator's skill and the quality of the (computer) user interface."
In contrast, partner-to-partner operation reduces the need for fine-grained planning and resource scheduling, according to Fong. Another advantage of partner-to-partner operation is that it "does not require the human to continuously engage in robot teleoperation or supervision," Fong said. "To improve system capability, increase flexibility and create synergy, human-robot dialogue needs to be richer and work both ways."
The system under development at NASA Ames will enable robots to interactively support one another so that they "can quickly jump in and help the other (human or robot) resolve issues as they arise," Fong continued. NASA Ames researchers will initially focus on developing robots that will operate more like human beings during construction and maintenance tasks.
Robots have worked as field geologists during previous field tests, but in the future, these human-like machines will also do non-scientific work. They will not only look for raw materials on planets, but also process those materials, according to researchers. Other robot jobs may include scouting, surveying, carrying equipment, inspecting and maintaining machines and habitats. Androids could well serve as field medics, monitoring, helping and transporting injured astronauts.
Human-robot teams must be reliable, survivable, reusable and low-cost, according to scientists, who hope to enable robots to receive general commands, such as, "inspect habitat." The human-like machines would then autonomously perform tasks and seek help only when they encounter problems they cannot solve by themselves.
"A challenge that scientists face is enabling robots to perform tasks on their own as much as possible, and giving these androids the ability to ask for human help and knowledge only when necessary," observed Fong.
Even so, robots will need to understand specific instructions in the same way that human beings often communicate them - in "spatial terms," according to scientists. "We (will) focus on spatial reasoning, so that humans and robots can converse more like humans do with each other, such as, "move that panel to my left," Fong explained.
Two robot characteristics are that these machines will communicate by speaking and gesturing. In places with little or no atmosphere, sound cannot travel well, and without radio communications, a gesture might be the most practical way for a robot and a human being to communicate, according to researchers. Trying to communicate in a very noisy area is another obvious reason to enable robots and humans to use gesturing to 'talk.'
"Because spacesuits are restrictive, they don't allow the astronaut to use devices such as small computers. Therefore, we are investigating the possibility of using pointing gestures to communicate with robots," Fong said.
Robots and human beings need to clearly converse about goals, abilities, plans and achievements, according to Fong. People and robots need to "collaborate to solve problems, especially when situations exceed (robot) autonomous capabilities," Fong explained.
According to Fong, human-robot communications need to take place on three levels: in a shared space, line-of-site (from a human being in a habitat to a robot outside on the planetary surface), over the horizon and even at interplanetary distances. To date, probably no one single robot system is capable of communicating on all these levels, Fong added.
Fong said that a robot that could work autonomously would enable astronauts to "focus on the task, instead of robot control and monitoring."
On the Net: