Human controllers will be involved at a higher level than in the past.
For instance, the robots are too complex to control every joint so a human may have to command it to turn a valve while the device figures which individual joints, like the elbow and the wrist, need to be engaged.
"The software is interpreting these high-level commands into low-level commands for how it will get that done," noted Connor.
And R.J. Linton, a Ph.D. candidate and member of the team from the Worcester Polytechnic Institute, said it's a much greater challenge to build the software for a humanoid robot, than for a four-legged or wheeled robot.
"The human form is bad in general," he explained. "We have an unstable gate. The humanoid shape is inherently unstable. You're always shifting weight to maintain balance. You have to teach a robot how to do that. Atlas has to always be kind of moving, just like a human."
Neuhaus said humanoid robots are in their infancy and called the challenge a good starting point for needed research.
"There are some technical challenges -- no one really felt it was possible," he said. "It involves solving a lot of problems at the same time. You have to solve the [artificial intelligence] challenge, embedded computers, actuation, algorithms for walking and balancing and manipulation. You have to fit all of that into the same research project. Researchers might work on individual parts but not the entire thing."
A big issue for a humanoid robot, for instance, is enabling it to have the hands do something while it's walking. Those two simple sounding tasks might involve 30 or 40 joints working simultaneously and in unison.
"We spend a few years of our lives learning how to control our [joints] and our sense of balance and pushing that task down to a reflex," said Neuhaus. "When a one-year-old is learning to walk, it's at the forefront of their cognition. They're really thinking hard about how to do it. When you're older, you can walk and think about other things at the same time."
Some teams are building their own robot and the software to run it, while others are developing software for the Boston Dynamics Atlas robot, a six-foot, two-legged, humanoid robot that, at this point, has the mobility and dexterity of a one-year-old child.
When the robotics challenge began, 100 teams registered and 26 qualified to compete. The number was thinned to 17 in June after the software was tested in simulation.
The 17 teams remaining are receiving financial backing from DARPA. The field will be narrowed to eight teams or less after the latest challenge.
According to Linton, a humanoid robot capable of assisting humans during a natural disaster is possible within the next 10 years.
Sharon Gaudin covers the Internet and Web 2.0, emerging technologies, and desktop and laptop chips for Computerworld. Follow Sharon on Twitter at @sgaudin, or subscribe to Sharon's RSS feed . Her email address is sgaudin@computerworld.com.