What if a robot had only one sensor?

You use your eyes to see. You use your hands to feel. You use your ears to hear.

Robots have sensors to do jobs like these. What if a robot had only one?

People find out about the world with their senses: sight, hearing, touch, taste and smell. Robots need to find out about the world too, and they do it with sensors. A camera is a sensor for light. A touch sensor notices a push. A microphone picks up sound.

In this thought experiment, Olly imagines building a robot with only one sensor. What would it be good at, and what might it miss? Choose a robot to find out.

A robot's sensors are its only link to its surroundings. Each type converts one physical quantity, light, force, sound, distance or chemistry, into a signal its computer can use, and each has conditions in which it gives little or misleading information.

This thought experiment strips a robot down to one sensor at a time, a camera and then touch, and reasons about the gaps. It then compares the result with a real robot designed to work far from any human help: NASA's Perseverance rover on Mars.

Real robots often have lots of sensors.

Some sensors are like eyes. Some are like hands. Some are like ears.

NASA's robots on Mars show how many senses a robot can have. NASA explains that cameras have given its Mars robots sight, robotic hands, arms and feet have given them touch, and chemical and mineral sensors have let them taste and smell rocks. Perseverance also carries microphones, so it can hear.

The rover has many cameras with different jobs. Six of them, called hazard cameras, watch for dangers in its path: four at the front and two at the back.

NASA states that robots have replicated much of the human sensory experience on Mars: cameras supply sight, robotic hands, arms and feet supply touch, and chemical and mineral sensors let scientists 'taste' and 'smell' the surface. Perseverance adds microphones; one hears the pop of its laser striking rock several feet away.

For driving, it carries six Hazard Detection Cameras (four in front, two at the rear) and two color stereo Navigation Cameras that help engineers navigate it, particularly when it drives autonomously.

Imagine a robot with only a camera. It can see, but it can't feel.

Now imagine a robot with only touch. It can feel a bump, but it can't see what's ahead.

Imagine a robot with only a camera. It might find a cup on a table, but it couldn't feel how hard it was squeezing, so it might drop the cup or crush it. In the dark, it might see very little.

Now imagine a robot with only touch. It might only know about a wall after bumping into it. A NASA scientist who planned a sensitive robot skin looked at it the other way round: he said seeing with cameras is not enough, and robots also need to feel what is around them.

A camera-only robot might locate objects well in good light but have no direct measure of contact force, grip or slippage, and little information in darkness or glare. A touch-only robot might detect obstacles only on contact, building a picture of its surroundings one collision at a time.

NASA Goddard technologist Vladimir Lumelsky argued in 2005 that vision is not enough. He was setting up a laboratory to develop a sensitive robot skin that would carry more than 1,000 infrared sensors to detect nearby objects and let the robot react within milliseconds. The gaps described above are reasoning, labeled as ideas in the lab.

Each sensor notices only one kind of thing.

A camera notices light. A touch sensor notices pushes. Using both, a robot knows more.

Every sensor measures just one kind of thing. A camera needs light to see. A touch sensor only works when something is touching it. A microphone hears sounds but can't tell colors.

When a robot uses several sensors together, one might fill the gaps of another, and engineers might check one sensor's information against another's. That might be one reason real robots often have many sensors, just as you use your eyes and your hands together.

Each sensor samples one physical channel with limited range, resolution and conditions of use. Combining channels, often called sensor fusion, might cover one sensor's blind spots with another and allow cross-checking, though it would need more hardware, power and computation.

In this lab, the specific weaknesses of single-sensor robots are reasoned examples rather than measurements of a particular machine, so they are labeled ideas. What is documented is that real exploration robots, such as Perseverance, carry many sensors of different kinds.

The Perseverance rover explores Mars. It has many cameras.

It also has microphones to hear, and an arm to touch rocks.

NASA's Perseverance rover drives across Mars, guided by engineers on Earth. Its two navigation cameras help it find a safe path, especially when it drives on its own. Its hazard cameras watch for big rocks, trenches and sand dunes in its path, in front and behind.

Some cameras help it drive, others take science pictures and help collect rock samples. Its microphones even heard the pop of its laser hitting a rock. Many sensors, each with its own job, let it do far more than one sensor could.

NASA describes Perseverance's cameras as split between engineering and science: several serve as eyes for driving, while others make science observations and help collect samples. Before the rover 'drives blind' along a planned path, its cameras help ensure that path is safe. Its microphones add sound, and robotic arms supply touch.

The design reflects the lab's main idea: a robot working far from help carries several overlapping senses rather than relying on one.