Imagine you're a platoon leader in Afghanistan, approaching the entrance to a dark cave and wondering whether enemy fighters are lurking inside.

You could send some men in, at great risk of their being shot. Or the Army could give you a robot, but they're expensive, relatively heavy and often stymied by difficult terrain.

But what if you could throw an object inside that would give you readings from multiple sensors?

That's the idea behind the Scout Ball, a softball-sized device being developed by MobileFusion, a South Side technology startup.

The Scout Ball is the brainchild of Abhishek Sharma, a native of New Delhi, India, and a graduate of Carnegie Mellon University's robotics program.

Mr. Sharma's expertise is in remote sensors, particularly the software needed to combine information from different kinds of sensors into one coherent wireless stream of data.

The Scout Ball is packed with video, audio, thermal imaging for night vision, seismic motion detectors and enough processing power to run a standard desktop computer.

The device is a ball because "a spherical shape is one of the most closed, tight shapes available, and it has the maximum tactile surface -- you can grab it and lob it. That was the motivation," said the 27-year-old computer scientist.

In a typical military scenario, the idea would be to throw one or more of the Scout Balls into an enclosed space and then see whether there is a human being inside and exactly where he is.

Each ball has a GPS receiver and an accelerometer, which can detect the position of the ball before it is thrown and then map where it has landed.

The information the sensors pick up is then transmitted wirelessly to a handheld receiver or a standard laptop. But it won't be like watching a surveillance tape, Mr. Sharma said. Instead, the goal is to send a simple signal -- flashing lights or beeping-- when it has found the target it is interested in.

There are two reasons for that, Mr. Sharma said.

First, the sensors would generate too much information to send it back in raw form with a standard wireless signal. Second, soldiers in dangerous situations need to spend as little time as possible processing information.

"When you're under stress," he said, "you don't want to look at data; you want answers."

Putting several sensors in the Scout Ball is based as much on psychology as it is on engineering, Mr. Sharma said.

If the device lands in the room of a building, for instance, the first impulse of a fighter probably would be to hide from it, he said.

Of course, the soldier could also shoot the Scout Ball, but if that happens, Mr. Sharma said, "that's good, because you're giving the best signature possible of where you are."

More likely, though, the soldier would try to avoid detection.

"If he knows it's a camera device, he'll just go out of the field of view. If you throw in an audio phone, he'll be quiet. If you have a thermal device, he'll wear [camouflage] clothes" that are specially designed to trap body heat.

Because the Scout Ball has all these sensors, plus a motion detector, it can combine partial information from each sensor and then try to figure out if a human being is in the room.

Mr. Sharma is the first to admit that his company is still trying to figure out the best way to do that quickly and accurately.

"Cumulatively, we have learned that we are not yet at the point where we can have full faith in the device," he said.

However, the company has received strong interest from the military, public safety equipment makers and others, he said, so it is forging ahead with the hope of creating an economical, reliable device that can save lives.

The real key to the Scout Ball's future success is that it is designed not to be perfect from the start, but to learn as it goes along.

Each time the device is employed, he said, the user will indicate whether it found what it was looking for, and both the hits and misses will become part of its evolving database.

"So let's say you throw it in a building, it gives you an alarm, you act on it, and you go in there and find a cat. Then you use a touch screen to say that was not a human, and it will keep on evolving and giving better estimates."

Eventually, Mr. Sharma sees the device being adapted for many other uses.

Police could use it when hostages are being held. Firefighters could use it to see if anyone is trapped in a burning building. Oil and gas companies could use it to see if a drill hole is near pay dirt.

One day, he said, MobileFusion hopes to create a modular device that different sensors could be plugged into, depending on the needs.

A team trying to rescue trapped miners would plug in audio and video sensors, for instance, while a gas exploration team could insert chemical sensors.

Mr. Sharma, who has been in Pittsburgh for six years, said with a self-deprecating chuckle that he was somewhat naive in deciding to develop a remote sensor for the U.S. military.

As a foreign citizen, he is limited in how much direct work he can do for the armed services. In fact, during recent tests of the Scout Ball at a Lockheed military research facility, he had to wear a different colored badge, and one employee was assigned to follow him wherever he went, including the restroom.

But Mr. Sharma, who is applying for a U.S. residency visa, said he values working for the military because "they will be our proving ground. They're fussy, they have hard problems to solve and they're less price-sensitive than private industry. And they understand that technology can work or sometimes not work the way they want."

Even working for the military, though, Mr. Sharma realizes the Scout Ball has to be reasonably priced. He hopes to get each unit down to about $3,000 or $4,000, which means 50 of them could be destroyed in action and still not equal the cost of the standard $250,000 life insurance policy on a U.S. soldier.

In the end, he said, he hopes to develop a technology that not only will save lives but will meet the difficult challenge of accomplishing what humans are able to do naturally.

"We're really good at working with partial information," he said, "because we superimpose on our partial measurements the prior information we've learned over the years."

By building a device that can learn as it goes, he hopes to duplicate that process.

"What we're saying to customers," he said, "is 'We'll give you the software, but you take the device, be friendly to it and make it your pet.'"