When we talk about the future of robotics, we immediately jump to humanoids.
Two arms. Two legs. A human-like body. General-purpose autonomy.
But I think that may be the wrong place to start.
If we want to build intelligent robots, we should first build something much smaller.
My recommendation would be to start with either an intelligent pendant-like device or a small robot dog.
Not because they are the final form of robotics.
Because they are excellent platforms for learning what it means for AI to exist in the physical world.
Start with a pendant
A pendant is interesting because it can stay close to a person.
It can listen to conversations.
It can understand context.
It can provide information.
It can act as an interface between humans and machines.
But the pendant has one major limitation:
it is attached to you.
Its microphones are constrained by where you are wearing it.
Its cameras have a limited perspective.
It cannot easily move toward a sound.
It cannot look behind you.
It cannot physically position its sensors where they would be most useful.
This is where a small mobile robot becomes much more interesting.
Give the AI a body that can move
Imagine a small robot dog that follows you around.
It doesn't need to carry heavy objects.
It doesn't need to run.
It doesn't need to perform acrobatics.
Its first job could simply be:
listen, see, understand, and move.
That sounds simple, but it creates an entirely different category of AI.
A smartphone has cameras and microphones.
A pendant has cameras and microphones.
But a small robot can choose where to put those sensors.
If someone is speaking from across the room, it can turn toward them.
If you ask it to look at something, it can move closer.
If there is an unusual sound, it can investigate.
If something is outside its camera's field of view, it can physically reposition itself.
The body becomes part of the intelligence.
Better ears than a pendant
This is one of the most interesting reasons to use a small robot dog.
Hearing is not simply about having a good microphone.
A robot can potentially have multiple microphones distributed around its body, allowing it to estimate the direction of a sound.
It could distinguish between:
"Someone is talking to me."
and
"There is something happening behind me."
It could turn its head toward the speaker.
It could move closer.
It could isolate voices from environmental noise.
It could potentially build a continuously updated understanding of the acoustic environment around it.
A pendant can listen.
A robot can orient itself toward what it is listening to.
That distinction matters.
Better eyes than a pendant
The same applies to vision.
A wearable device might have a camera pointed roughly where the person is looking.
A robot has another possibility:
it can move its camera.
It could look up.
Look down.
Turn around.
Move closer.
Move farther away.
Look underneath something.
Look at the person speaking.
Look at an object the person just mentioned.
The camera is no longer a passive sensor.
It becomes an actively controlled sensor.
That is a profound change.
The AI doesn't just receive the world.
It can decide how to observe the world.
This is the beginning of embodied intelligence
This is where I think small robots become much more important than they initially appear.
An AI model running on a server can understand language.
It can reason about images.
It can generate plans.
But it doesn't experience the physical consequences of its decisions.
A small robot does.
It sees something.
It makes a decision.
It moves.
Its camera changes.
Its microphones receive different information.
It bumps into something.
It succeeds.
It fails.
It tries again.
The intelligence exists inside a continuous loop:
Perception → Understanding → Decision → Movement → New perception
That loop is the foundation of embodied intelligence.
Don't make the robot complicated
The temptation will be to add everything.
More motors.
More actuators.
More degrees of freedom.
More sensors.
More complicated mechanical systems.
I would do the opposite.
Make the physical robot extremely simple.
The intelligence should come from the software.
A small four-legged platform could be enough.
A camera.
Several microphones.
Depth sensing if necessary.
A speaker.
Basic onboard compute.
Wireless connectivity.
Battery.
And an intelligent model controlling the system.
The robot doesn't need to be mechanically impressive.
It needs to be useful as an intelligent sensor platform.
The robot becomes an AI peripheral
We normally think of peripherals as things connected to computers.
Keyboard.
Mouse.
Camera.
Microphone.
Speaker.
Perhaps the small robot is a new kind of peripheral.
An AI peripheral for the physical world.
Instead of bringing the physical world into the computer through a fixed camera and microphone, the robot can move through the environment and bring information back.
The AI can effectively say:
I need to see that.
And the robot moves.
I need to hear where that sound came from.
And the robot turns.
I need another perspective.
And the robot changes position.
This makes the physical body an extension of the AI's sensory system.
And this is why the robot dog may be better than the pendant
The pendant is still an excellent starting point.
It is cheaper.
It is easier to wear.
It is socially less intrusive.
It can become an always-available interface.
But a small robot dog has something a pendant cannot easily provide:
mobility.
It can leave your immediate position.
It can observe independently.
It can return.
It can follow.
It can investigate.
It can position itself for better sensing.
That makes it much more interesting as a research platform for physical intelligence.
The first killer application doesn't have to be physical labor
This is another important distinction.
We often justify robots by asking:
"What physical task can the robot automate?"
But the first killer application for an intelligent mobile robot might simply be information.
Imagine having a small robot nearby that can:
- listen to a conversation
- answer questions about what it sees
- find something in a room
- observe an experiment
- monitor equipment
- inspect an environment
- recognize unusual sounds
- follow a person
- provide another camera angle
- remind you about something it observed earlier
The robot isn't necessarily replacing a worker.
It is giving a person another set of eyes and ears.
And that may be a much easier place to start.
Small robots can become incredibly powerful experiments
The biggest advantage of starting small is iteration.
You can build one.
Break it.
Change the microphones.
Change the camera arrangement.
Change the model.
Change the control system.
Try another task.
Then build ten.
The hardware becomes a laboratory for experimenting with intelligence.
And once the intelligence works, the physical platform can evolve.
The small robot might eventually manipulate objects.
It might learn more complex behaviors.
It might connect to other robots.
It might become a specialized worker.
But none of that needs to happen on day one.
Don't start with the perfect robot
This is perhaps the most important lesson.
We shouldn't wait until we can build the perfect humanoid.
We should build small embodied systems today.
A pendant can teach us how AI interacts with a person.
A small robot dog can teach us how AI interacts with a physical environment.
Both can collect real-world feedback.
Both can expose the limitations of today's models.
Both can help us understand what sensors matter.
And both can become stepping stones toward much more capable machines.
The future of robotics may eventually contain humanoids.
But we don't have to begin there.
We can begin with something sitting on a desk.
Or something quietly walking beside us.
A tiny robot with better ears, better eyes, and just enough intelligence to understand why it is looking and listening.
Because the first truly useful robot may not be the one that can do the most work.
It may be the one that can perceive the world for us.
Top comments (0)