ASIMO

This prototype mobile controller was seen being used at ASIMO's announcement demonstration. It consists of a computer running Windows attached at the waist and a game console controller used to direct its actions.

Variations

The color variations include gray, red, green, and blue. There is also a unique one that has an entirely different head, featuring it's large eyes and a smiling mouth that are usually hidden behind it's visor, but now on full display. This new design was shown off during the the rentals announcement.

Honda International Technical School (HITS), with their robot "Firstep", a modified ASIMO 2000 prototype, competing at RoboCup Japan Open 2003 Niigata. They would then attend RoboCup World Tournament 2003 Italy, and come in first, winning the Louis Vuitton Cup.

This was the first time a team outside of Honda R&D were allowed to work on the ASIMO platform. It was designed to kick a soccer ball by hitting it through it's walking gait.

Firstep's height is 125 cm and weights 50 kg. Replacing ASIMO's head was a custom head with a screen for a face, displaying animated expressions like the one above. There was also a custom enclosure for the backpack with upgraded wireless communication antennas, but was reverted back to the original for Italy.

2001 (Software Revision)

On November 12th, 2001, an announcement of a new ASIMO model was made. However it was clear that this was more of a refinement of ASIMO 2000. With one of the major improvements being a reduction in startup time from 40 minutes to 4 minutes. The time reduction comes from automating a lot of the manual tasks previously needed during startup. Another would be the improvement of the i-WALK technology, adding the capability of walking in three dimensions. This added the ability to climb stairs and slopes. Yes, ASIMO 2000 had the ability to climb stairs, however every stair had to be within millimeter spec to each other, leaving no room for error. If one step was a little higher than the other, ASIMO could not adapt to it, leading to it falling over.

ASIMO now had a better sense of it's surroundings, with the addition of a CCD camera positioned between the legs to track markers placed on the ground, making positioning easier and more accurate. It was also now possible to process voice commands and detect direction of sound sources.

Two last improvements, now there no need for workstation PC to handle offline tasks, as software and hardware had progressed to where only a laptop was necessary. And lastly, the cable conduit that ran from the backpack to the back of the legs has been removed.

2002 (Sortware Revision)

On December 5th, 2002, ASIMO was upgraded to track and recognize faces, and follow human gestures, like moving to where someone is pointing to or halting in place. It could also recognize the posture of a person and recognize its environment. Also, with added internet connectivity, it was able to provide news and weather updates.

The overall system architecture as been redeveloped. For preplanned routes, ASIMO interfaces with a external map management system located offline. It can also trigger dialogue and other tasks at given locations. However onboard, 3 separate processors handle image processing, speech recognition and synthesis, and control and planning respectively. A frame grabber is connected to the vision computer via a PCI bus.

The image pipeline consisted of stereo SONY XC-ES alternate 1/3" color CCDs, with a depth map generated using a SAD (Sum of Absolute values of Differences) based method on a Mobile Pentium III-M 1.2GHz. 3D objects and moving objects are then extracted by taking images from the frame grabber and processing them on the vision computer, sending necessary data to the navigation and interaction systems. In order to calculate distance and direction of objects in a scene, the image had to be synchronized with every joint angle, calculating the camera pose. With this, detecting multiple moving objects in an image was now possible. ASIMO stores the extracted objects' bounding boxes in its local map, necessary for obstacle avoidance for planning detours.

To track humans, they need to be extracted from the image. An optical flow based algorithm separates the foreground from the image, then Snake algorithm extracts the human shapes, separating multiple people in a scene. Face detection then uses skin color to identify face contours.

There are two gesture recognition systems. The first is 2D, recognizing gestures including handshake, hand circling, bye-bye, hand swing, high-hand and come here call. And the second is 3D, recognizing pointing gestures, using the head and hand positional relationship and the depth map to then move to the desired location.

ASIMO is capable of storing and identifying about 10 people. The database of faces and the details along with them are stored on the external management system.

The planning system is behavior based, allowing flexibility and modularity, making reactive in rapidly changing environments.

It was also possible for ASIMO to be connected via direct attach cable for video output of it's vision. It seems to use something like a D-Sub 15 pin display cable.

For a more in depth analysis on how each of the onboard systems work together, check out this paper.

As of February 21st, 2002 in an article by Forbes, it was revealed that 20 ASIMO units had been built up to that point, all hand assembled in their research facility in Wako, Japan. It was also revealed that a single unit costs just under $1 million to make.