Does a robot process "sensory data" or "sensor data"?

What is the best name for the data a robot acquires from its environment? Examples include camera images, laser range measurements, acceleration measurements, sound, the force it is exerting on a grasped object, etc.

I think "sensory data" implies a parallelism with human senses, which is often appropriate, but not always - e.g. humans can't accurately measure the distance of faraway objects like a laser rangefinder can, and we don't have GPS.

"Sensor data" loses this connotation, but it also becomes too broad. For example, a robot might have a battery life sensor, which delivers sensor data, but not the kind of sensor data we are trying to describe here.

My limited-but-real study in robotics and cybernetics suggests that ‘sensor data’ or ‘sensor input’ would be the conventional terminology, as supported by this educational page at (hooray!) robots.com.

This is not a hard-and-fast rule, however, as I will try to show, while also digging a little into what might really be at stake in the human comparison that you correctly note.

I think that you are wise to eschew ‘sensory’, roughly for the human-comparison reason that you give. The term seems to suggest a close parallel with biological sensory systems, an idea that is misleading on several grounds, not least of which is that our sensory mechanisms are integrated into a pretty established composite system (even if researchers like Kevin Warwick are working hard to cyborgise us), while any given robot’s array (let us say) of sensing capabilities is arbitrary: we can change it at will. On balance I would argue that your distinction here is definitely useful.

Some do use ‘sensory’ in a robotics context, however. In ‘Acquiring Observation Models through Reverse Plan Monitoring’, Chernova et al. (p8) explain that ‘The robot walks using odometry (and not sensory data) to traverse the preset distance,’, i.e. the motion sensing that you mention... which brings me to another main point, because Chernova et al. are not really taking about input from the environment, unless you draw the system boundary around the robot’s processor, and regard its sensors as existing in the environment... which you might...

If you see the robot as a unit, however, then Chernova et al. are talking about a preprogrammed routine to (attempt to) travel a certain distance in a certain fashion, regardless (as they say) of data sensed from the environment.

My sense of the robotics discourse is that the term ‘sensor data’ or ‘sensor input’ exists primarily to distinguish input that the robot collects dynamically in real time (on its travels, or whatever) from its environment (as you say), from input that it receives from its human users, via either its programming or arbitrary interventions (e.g. the campaign to nudge the robot lander Philae back to life after it landed on the wrong part of its target comet and lost power, or [I would argue] the preprogrammed motion sensing strategy described above). If this is so, then a preoccupation with the distinction with human systems might be a real red herring.

It also seems to be entirely normal for people to anthropomorphise robots (and other machines, like grumpy cars and computers) to some extent anyway. You will probably be aware of lots of research on this kind of thing. I just found a lovely paper on ‘How People Anthropomorphize Robots’ (Fussell et al.), analysing reactions to varying ‘behaviours’ of a robot ‘interviewer’.

Taking this human tendency judiciously into account, I see nothing wrong with allowing for figurative parallels such as conceiving of the battery life sensor as doing a job analogous to hunger. The University of Washington’s Sensor Systems Laboratory specifically explains one example in terms like this: ‘Our mobile manipulation plaform Marvin [...] feeds itself by “smelling” its food, electricity,’ and (not uniquely among robots, as you will know) plugging itself in when it is ‘hungry’. Marvin senses the need for ‘food’ (arguably making the battery part of the environment, in a sense), and then also senses its location. The SSL cheerfully and harmlessly uses the biological metaphor to get this across.

Similarly the function of a robot’s GPS access is basically a lot like a human looking at a map (or a GPS, come to that). The metaphors seem harmless. If that is so, the trick is not to play-down an anthropomorphic tendency but to organise your texts such that specialists like yourself can recognise them as essentially true and not technically misleading, while (relative) laybeings can get the drift of what you are talking about.

Broadly, then, I have satisfied myself that it is impossible to argue that either ‘sensor’ or ‘sensory’ is definitively correct, but I can see simple advantage in adopting ‘sensor data’ or ‘sensor input’. That said, it could still make sense to describe a robot as having particular ‘sensory capabilities’. Some of my reasons for coming to this whole view are kind of reverse-engineered from less technical work on anthropomorphism, and your view on all of that might depend on your relationship with the social sciences.