All of our actions are created by first obtaining information using our five senses,
comparing it with past experiences and analyzing it, and deciding what to do.
This process is repeated many times as we grow up so it becomes habitual,
something we do without thinking.

For example, when children hold an egg they sometimes squeeze too hard and break it,
but adults have repeated the experience multiple times so can hold it without thinking.
This is called unconscious cognition.
For robots to replicate this behavior they need to be able to sense things very accurately,
analyze quickly and precisely, make predictions, and act and talk in an appropriate way.
Front-end AI makes it possible for PALRO to have this kind of intelligence.

The front-end AI outputs conversation and actions through precise control,
so PALRO can act in a way that feels natural.
When things require deeper thought or more complicated analysis,
PALRO can use the strong intelligence of the cloud AI to respond.

At FUJISOFT we have accumulated knowledge from robot control research in collaboration with
industry, government and academia, and the ability to put this into action.
We used FUJISOFT technology to develop the hardware required for the control software, and this
became PALRO.
We achieved a balance between the software and hardware using robotics engineering technology
such as shock resistance, drive durability, heat dissipation, sensing accuracy, and analysis accuracy,
and created a front-end AI that makes PALRO highly functional and reliable.


PALRO looks at you when it speaks
Humans have a horizontal field of vision of around 200°, and a vertical field of vision of around 125°.
PALRO has a horizontal field of vision of 52° and a vertical field of vision of 40°.
To compensate for this more limited field of vision PALRO maps the surrounding area in 3D,
and analyzes the position of the people it detects using its microphones and ultrasonic sensor.
It combines the direction of sound and the distance from the person to naturally
turn to the person and center them in its field of vision.

Recognizing unclear words
In Japanese the difference between positive and negative sentences can be hard to discern.
For example, when using the verb "suru" (to do), a positive sentence ends "shimasu" and a
negative one ends "shimasen". This simple difference completely changes the
meaning of the sentence, and even people sometimes mishear it.
PALRO's speech recognition analyzes to the last sound in each word and provides PALRO
with the correct answer.
PALRO also simultaneously analyzes the actions of the person it is talking to.
Nodding is positive, and shaking the head is negative.
The answer appears naturally through actions and the tone of the conversation.
PALRO combines speech recognition and image processing to understand
what the person means.

PALRO can catch immediate answers too
When people know how they will answer a question they tend to respond as soon as the
other person has finished speaking.
PALRO prepares the most appropriate speech recognition data (standby words),
and waits to hear the person's voice while eliminating background noise.
This allows it to clearly hear the person's voice, and it recognizes the speech with a
high degree of accuracy.
PALRO's front-end AI also analyzes the meaning of the words, thinks, and makes it
possible to have conversations like a person.

Natural replies
When you call PALRO's name, it responds within 0.4 seconds.
If you ask PALRO about the weather, it will respond within 0.4 seconds.
One of the elements that makes communication feel natural is response time.
PALRO processes everything through the CPU in its housing, and responds after a gap
appropriate to the conversation (0.4 to 0.9 seconds) with a reply or action to create natural conversations.
It also accesses the cloud in the background to get detailed information required for the conversation.


When PALRO sees you, it calls your name within 0.08 seconds
PALRO centers people's faces in its field of vision, continuously predicts actions accurately,
and starts searching as soon as it sees a face.
This takes 0.08 seconds. The combined power of the front-end AI allows it to quickly
identify individuals and say "Hello (name)".