Because recognition targets, vehicle speeds, and traffic rules differ greatly between driving on roadways and on public open spaces, the map-less cooperative driving functions are configured separately for each scene.

The system is capable of real-time recognition of not only visible road surface markings such as lane markings and stop lines but also road structures such as lanes that people presume while navigating. Unlike maps, the road structures recognized by the system change every moment. The system generates action plans which accommodate such changes and controls the vehicle to make right/left turns with a stable steering angle.

Three-dimensional point clouds are generated from the parallax of the two lenses of a stereo camera, and the grid-based analysis of such point clouds enables the recognition of 3D objects which could be possible obstacles to CiKoMa.

If the height of the obstacle exceeds a predetermined value, the system determines it as an impassable area and finds passable areas quickly. Then, the system generates a rough outline of the shortest route to the target spot on a real-time basis, and CiKoMa moves along that route in a trajectory that smoothly avoids obstacles.

WaPOCHI continues following the user, who was authenticated by palm vein authentication, using its cameras by capturing and remembering the appearance features of the user.

WaPOCHI identifies the user by scanning the people who are in the vicinity of the last spot where the user was recognized, and whose characteristics captured by the cameras are most similar to the memorized characteristics of the user.

Let’s imagine the future where self-driving mobility products are widespread and a situation occurs where the user wants to inform the mobility product of a change in the meeting point and/or the destination.

If the user has to use a smartphone app and type in the new location to notify the mobility product, it would not only be inconvenient, but in some cases, too slow to deal with last-minute changes.

If the mobility product is capable of understanding human words and gestures, just as people communicate with each other, the user and the mobility product can smoothly communicate their intentions. With Honda CI, we are taking on challenges to realize “mobility at the will of the user,” which will not ask human users to accommodate the needs of mobility products but enjoy mobility based on mutual understanding between machines and people.

CiKoMa first moves to the approximate location where the user is located based on GPS information, and then searches for the user by applying image recognition using images captured by its cameras. If there are multiple people in the vicinity, CiKoMa determines the characteristic differences among the candidates and automatically calculates which questions about their characteristics will help in identifying the user in the shortest amount of time. Then CiKoMa identifies the user by asking these questions.

In addition to words, people often indicate the direction they want to go by pointing their finger in that direction. For example, if there are multiple vending machines and the user points to one of them and says, "Go to the vending machine over there," CiKoMa first extracts the keyword "vending machine" from the instruction. CiKoMa then uses its image recognition function to gain three-dimensional understanding of the pointing gesture and estimates the direction the user is intending to point. By understanding this information in an integrated manner, CiKoMa smoothly identifies the user's intended destination, just as if the user were interacting with another human being.