Teleoperation

On 11 March 2011, an earthquake and tsunami struck the Fukushima Daiichi nuclear power plant in Japan. In the days that followed, some of the first responders who entered the reactor buildings were not human — they were tracked robots, controlled by operators sitting safely outside the radiation zone, using joysticks and camera feeds to guide the machines through rubble and read instrument panels. The operators' hands never entered the building. Their agency did.

This is teleoperation — controlling a robot from a distance, with a human making the decisions in real time. The word fuses the Greek tele (far) and operation. It is distinct from both full autonomy (where the robot decides on its own) and simple remote control (which implies the human is just issuing movement commands to a dumb machine). In teleoperation, the human provides intelligence; the robot provides physical presence somewhere the human cannot or should not be.

The anatomy of a teleoperation system

A teleoperation system has three essential components. First, a command interface — whatever the human operator uses to express their intent. This might be a joystick, a haptic glove, a pair of tracked motion-capture controllers, or even a full exoskeleton that the operator wears and moves naturally. Second, a communication link — the channel between operator and robot. This might be a radio link, a fibre-optic cable, or an internet connection. Third, a feedback channel — information flowing back from the robot to the operator, most critically video but increasingly also force and tactile feedback (so the operator can feel what the robot is touching).

Time delay: the central engineering challenge

When the distance is small — a surgeon at a console controlling arms across a sterile field, as in the da Vinci Surgical System — the communication delay is essentially zero and teleoperation is smooth and intuitive. When the distance is large, delay becomes the defining constraint. Controlling a robot on the surface of Mars from Earth involves a one-way signal travel time of between 3 and 22 minutes, depending on orbital positions. You cannot drive a Mars rover in real time; every command requires a wait-and-see cycle. NASA's Curiosity and Perseverance rovers handle this by executing pre-planned sequences of moves, with human oversight happening in discrete review sessions.

Even at terrestrial distances, a delay of more than about 200 milliseconds makes teleoperation noticeably awkward, and delays over 500 milliseconds make it genuinely dangerous. Techniques like predictive displays (showing the operator a simulation of where the robot will be based on their current input, rather than waiting for the delayed camera feed) help bridge the gap.

Teleoperation versus autonomy

The two are not mutually exclusive. Most sophisticated field robots combine them: the robot handles low-level navigation autonomously while a human operator makes higher-level decisions, designates targets, and intervenes when the autonomous behaviour fails. This model — sometimes called supervised autonomy — captures the complementary strengths of both.

As robots become more capable, does it become less ethical to send them into danger — or does that question only apply to robots that can suffer?