PALRO finds people quickly and faces them

Facing the other person when talking is something people have been doing unconsciously since before history began.
If the person you are talking to looks elsewhere, you may wonder whether they are actually listening.
PALRO finds people quickly, and faces them properly.
It uses many things to make overall judgments, such as who the person is, who is talking, and whether the conversation is related to PALRO.

PALRO looks at you immediately

PALRO maps its surroundings in 3D, and is always aware of changes in its surrounding
It captures 3D mapping data that changes dynamically and moving bodies, and combines this with its predicted position coordinates to find people.

PALRO identifies you using extremely dense mapping
PALRO can remember more than 100 faces

When it finds a face, it creates a dense map of its features, such as the eyes, nose, mouth and profile. It uses this to quickly identify the person and match a name to the face from its database of friends.
If it is someone PALRO knows, PALRO will call their name and start talking about lots of different topics.


PALRO will always hear your voice

PALRO continuously analyzes surrounding sounds so that it is always ready to talk.

Dissecting audio for automatic trailing orientation control
PALRO uses 4 microphones to dissect and analyze surrounding sounds from all directions.
If it identifi es someone facing and talking to it, it automatically orients itself in that direction. It then only obtains the sounds coming from that direction, and tries to detect speech data.

Speech detection
PALRO uses frequency and audio levels to detect the parts of the audio data that have a high chance of being speech, and fi nds the start and end of the speech extract.
PALRO also sets up margins before and after the extract so it does not miss any weaker speech at the beginning and end, and uses this to fi nd segments to be used for speech recognition.

Speech recognition
PALRO uses fragments of waveforms cut from the speech recognition segment and compares them with model data of the vowel and consonant sounds used in Japanese.
It fi nds similarities to see if the waveform fragments match Japanese sounds.
These waveform fragments are processed individually, and incorrectly identifi ed data is fi ltered out to create the fi nal character data.

Noise reduction
Noise generated by the actuators when PALRO is moving, the ultrasonic noise created by the ultrasonic sensor and other noise, and environmental sounds can reduce the accuracy of speech recognition.
PALRO eliminates noise in real time to match its surrounding environment.

An algorithm that automatically rejects unknown words
Generally, PALRO uses the speech data found using its speech recognition and does its best to match it to words that it already knows.
This means that it may incorrectly identify the various sounds generated around it as a word it already knows.
PALRO has an algorithm that ignores words it does not know, which allows it to always identify speech without needing to press a switch or button.