Humanoid robotic shapes are exciting, certainly, but not necessarily superior to others. Superiority in one aspect does not translate to overall superiority.

Humanoids could be most useful as scientific research platforms for research into human perception, behavior and entertainment.

The potential for a robotics industry is very big. But I'd like to stress that it is the "robot technology" industry, or RT, industry. Just as the IT industry is not limited to PCs, the RT industry would encompass all robotics related technologies.

Major industry
For example, a surveillance camera that can automatically track suspicious persons will incorporate perception and mobile components. On that basis, the robotic industry will be a major industry — comparable, perhaps, to the automobile industry.

The markets are not just there. They have to be discovered. Most people don't even realize what would be possible through robotics technology.

We expect to see the emergence of many small, entrepreneurial, venture-backed companies. Many of them will fail, but some of the successes will become the Microsofts or Intels of this industry. But we're not there yet.

Even current technologycan find markets. It's a matter of finding out where the technology is needed and applying it at the appropriate level. Not too expensively, not too sloppily, just appropriately. Technology is a necessary but not sufficient condition for success. Technology probably accounts for 30 percent of market success. The rest is marketing, customer satisfaction and everything else.

Robotics is the integration of various disciplines; software, hardware, mechanics, devices, electronics, applications ideas, marketing, everything is involved.

Faster processors
For next-generation robotics we need faster processors; many hundreds of times faster, so that we can do real-time vision systems, sophisticated real-time decision-making and sensor reading. That's really important.

Second, we need innovation in material science. We need soft structures; we need artificial muscles instead of motors and gears.

Third, AI and perception research is getting better, but still we need advances in learning adaptation. Manufactured robots have to adapt to the environment by learning. And, they have to be able to operate based on built-in overall instruction with details filled in over time.

It is a combination of top-level instruction and low-level learning that leads to instruction-based adaptation. For example, suppose you have a robot that just goes from here and there. That's the instruction. But the robot has to be able to cope with many different conditions along the way — a 5-cm difference in floor height, for example. Programming for all possible conditions is not possible. So, a human should be able to give high-level instruction with the robot achieving that instruction by learning and adapting. That's the kind of technology that we need.

It's not unlike a soccer coach who instructs each player on his or her role, but leaves the on-field execution to the player.

Judgment and final decision-making should be left to human wisdom. I don't think we should expect to see robots with human-level intelligence. Of course, in the end, it is cheaper to hire human beings than building robots. Personally, I much prefer human colleagues.