HEARBO dancing, and sensor location breakdown (via HONDA)
Whatever futuristic world you wish to build with robots, one thing must be certain; robots of the future will need highly sophisticated ways of communicating and understanding humans. For this to happen, the field of computational auditory scene analysis will need to be greatly improved upon.
That's the goal behind Honda’s HEARing roBOt, aka HEARBO. This sleek looking bot has an elaborate system for detecting sounds. Using its 8 microphones implanted in its head, it can interpret 4 distinct, but simultaneous sounds, and pin point their source within 1 degree of accuracy. It does this by using sound source localization (SSL). Meaning, that instead of filtering out sounds, it locates the source and then analyzes the sounds individually to determine what the robot interprets as “noise” and what could be something truly meaningful like a baby crying. Microphones implanted within HEARBO’s body cancels out the sound made by its 17 motors when the robot dances hardcore to music or turns its head to locate sounds.
This technique of localization, separation and recognition is called beamforming. At HRI-JP, researchers Kazuhiro Nakadai and Kelsuke Nakamura developed a more refined version of the beamforming algorithm called HARK (HRI-JP Audition for Robots with Kyoto University).
The researchers have taught HEARBO how to distinguish specific patterns of sounds. So, it can distinguish a singing human from a talking one in the same environment.
Theoretically, its 8 microphones could triangulate up to 7 sounds so researchers will attempt to make this possible soon. The HARK program and programs alike must be refined before bots can tackle the noisy environments humans usually work in. HARK’s and HEARBO’s advancements towards this goal were demonstrated at the IROS 2012 conference.
Cabe