Aibo may soon have a master it can relate to. Sony has taken the wraps off its latest prototype humanoid robot, the SDR-4X, and said that it could be on the market by the end of this year.
The SDR-4X is a successor to the SDR-3X prototype that was unveiled in late 2000 at Robodex, the world's first exhibition of entertainment robots. Now with Robodex 2002 a week away, engineers from Sony's Digital Creators Laboratory in Tokyo are showing off their latest prototype and the advances they have made in robotic technology in the last 16 months.
The researchers face the problem of making the technology small enough to fit inside the SDR-4X, which is just 580 millimeters tall -- around a third of the size of Honda Motor Co. Ltd.'s well-known Asimo humanoid robot -- and weighs 6.5 kilograms. Powered by a pair of RISC (reduced instruction set computer) processors, the SDR-4X is capable of 38 separate degrees of movement, a significant jump from the 24 of the previous version.
Among the many improvements in the new version, two areas are worthy of note: a real-time adaptive motion control system and richer communication technology including facial recognition and complex speech recognition.
The motion control system enables SDR-4X to go places and do things its predecessor couldn't, such as balance on a moving platform and walk on carpet or Japanese tatami mat floors. While they may not seem difficult, for a robot they are major tasks, said Yoshihiro Kuroki, a senior manager at the Digital Creators Laboratory and the original brains behind the SDR project.
"Most small humanoids have a problem to adapt to such a floor," he said. "A big humanoid has weight so it is easy to walk on this kind of floor but a small humanoid does not have that advantage." Considering the SDR-4X's height and weight, he said it is equivalent to a robot the same size as an adult trying to walk on a carpet around 1.5 centimeters deep and keep its balance.
The development of a real-time motion-control system makes all of this possible. The system dynamically adapts to changing terrain and helps SDR-4X keep its balance. "We developed four small sensors on each foot and a gyroscope in the body," he said. "That is a very new thing so now the SDR has the ability to adapt to irregular terrain."
A pair of digital cameras have also given SDR-4X stereoscopic vision, allowing it to negotiate around objects that are in its way. The previous model had a single camera and so was unable to perceive depth.
The cameras also helped engineers in tackling the second major challenge: improving the interaction between the robot and humans. SDR-4X has the ability to remember the faces of 10 people and can even recognize emotions from facial features. Improved voice and speech recognition technology has provided it the ability to recognize people and also understand continuous speech, rather than the single words that the SDR-3X or Aibo robots can manage.
"For the intelligence part, the SDR-3X used Aibo technology which is discrete word recognition and color recognition technology," explained Masahiro Fujita, a senior manager in charge of interaction technologies used in the SDR-4X. "Today, we have developed three new recognition technologies -- face recognition, continuous speech recognition and short and long term memory."
The speech recognition system, while vastly improved upon from the previous model, is still a work in progress, said Fujita, something underlined by the difficulty the SDR-4X had in recognizing some continuous speech during Tuesday's demonstration. Background noise, said Fujita, was proving particularly difficult for the engineers to tackle.
"For speech recognition, there are several kinds of noise," he said. "There is white noise, for robots there is motor noise, and also background noise. For the first two, we can reduce the noise level using multiple microphones." The SDR-4X has a total of seven microphones to enable it to work out the direction from which sound came and better hear it.
"But if two people speak to the robot simultaneously, we have the so-called cocktail party effect. We have to focus on the particular person and this is a problem. Using the seven or six microphones we will make an effort to improve the recognition rate but, basically, this is still a problem."
It is one of several issues that engineers will have to tackle before Sony can begin selling the product, said Kuroki. The company needs to work on increasing reliability of the hardware and software but, should all go according to plan, the first commercial version of the SDR could be on the market before the end of the year. Sony wouldn't comment on the price, but it is expected initially to be considerably more than the company's Aibo robots.
More immediately, the SDR-4X will be unveiled to the public at the Robodex 2002 exhibition next week in Yokohama, Japan. With the theme of "robots as partners," the exhibition will highlight the latest in robot technology from leading robot researchers.