Long-distance robots take to the treadmill
In the exhibition area at the 2015 DARPA Robotics Challenge Finals, a pair of humanoid prototypes built by SRI and Sandia National Laboratories hit the treadmills to demonstrate new technology that allows robots to operate longer on a single battery charge.
Many modern robots can walk, or, like the Boston Dynamics' Cheetah, run, but what hampers most of them is a lack of stamina. As such, the 2015 DRC was designed to encourage improvements in the endurance of autonomous robots, with no external power allowed for competing robots.
According to Rich Mahoney, director of SRI Robotics, robots aren't like electric cars. Batteries are not the problem and standard lithium ion units are still adequate for most needs. The challenge lies in improving efficiency through both mechanical hardware and software.
To improve endurance, it isn't simply a matter of seeing how long a battery hooked to a system of servos can hold out. It's a question of how efficient a complete walking system can be constructed that can not only walk, but do so over enough different terrain to be practical.
The first robot in the demonstration was developed by Sandia National Laboratories as part of a DARPA project to produce more energy-efficient robots. Called the Sandia Transmission Efficient Prototype Promoting Research (STEPPR), it was a research platform designed as a lab bench where engineers could try different types of joints.
This was succeeded by the Walking Anthropomorphic Novelly Driven Efficient Robot for Emergency Response (WANDERER), with which Sandia Labs took the lessons learned with STEPPR and placed them in a more efficient package. Using brushless DC motors with very high power-to-weight ratios, and low-ratio transmissions combined with low friction joints and support elements to improve efficiency. Another factor to improve battery life is making things quieter, since, as Sandia points out, noise is lost energy.
The second robot in the endurance demonstration was SRI's PROXI, which made its public debut at the DRC. It was billed as 20 times more efficient than current humanoid robots and capable of up to eight hours of walking on a single charge. Sporting everything from spring-mounted ankles to torque-sensing joints, PROXI, which is just a torso and legs in its basic form, weighs 100 kg (220 lb), holds a 19 kg (42 lb), 2.2 kWh battery, and can be fitted with arms and a sensor-packed head.
According to Mahoney, the real advances are in new motors, power algorithms, materials, and generally making making the systems move and use energy more efficiently. PROXI is designed as open-source research robot, though parts of it have practical applications.
Mahoney says that humanoid robots are important because they can work in a human environment, and are easier to control because their body is intuitively like a person's. He went on to say that he sees robots as both the new industrial revolution and a sort of material internet with the ability to bring currently scarce services to people all over the world who currently cannot enjoy them. He predicts that robots will enter everyday life inside of five years, though not in humanoid the form, but more likely as components in factories, hospitals, restaurants, homes, and businesses.
“SRI, which is at the leading edge of robotic manipulation developments, is applying new component technology to expand the design space for robotics,” says Mahoney. “We are laser-focused on three key areas that will enable a new generation of solutions – low-cost, high-performance manipulators, wearable robotics, and versatile high-rate micro-manipulation. The DARPA Challenge, which convenes some of the world’s most advanced robotics R&D organizations, is the ideal opportunity to show how we are addressing real problems with new applications."
So how did the two robots go on the treadmill? They covered a 2.1 km (1.3 mi) over a period of 2.5 hours. While this wasn't set up as a race against the clock, it does show that there's some way to go before humanoid robots reach human levels of endurance.