Page 149 - Handbook of Biomechatronics

146 Domen Novak

forward on its own, thus requiring the user to only input actions if they want
to change the wheelchair’s behavior. Obstacle avoidance is achieved by
means of cameras and sonar sensors attached to the wheelchair; these sensors
constantly scan the area around the wheelchair, creating an “occupancy
grid” of nearby obstacles. If an obstacle is detected partially in the wheel-
chair’s path, it is treated as a repeller in the occupancy grid, causing the
wheelchair to automatically swerve to avoid it and then continue on its orig-
inal path. However, if an obstacle is directly in front of the wheelchair, the
wheelchair will slow down and smoothly stop in front of it, then remain
stationary until the user executes a turn command via the BCI. This allows
the user to “dock” with an object of interest (e.g., a table or sink) by aiming
the wheelchair directly for it. Such a shared control paradigm successfully
combines the intelligence and desires of the user with the precision of the
machine, allowing experienced unimpaired users to complete tasks using
the BCI approximately as fast as using a two-button manual input. We
believe that such shared control, where users give high-level commands
through a BCI and the machine takes care of low-level details, represents
the future of practical BCI control and will be adopted by a broad range
of applications.

2.2 Control of Mobile Robots and Virtual Avatars

The same principles described in the previous section can be used to control
not only wheelchairs, but also all other types of mobile robots and even ava-
tars in virtual environments. For example, in a classic study by Milla ´n et al.
(2004), two participants were taught to steer a mobile robot through mul-
tiple rooms using motor and mental imagery. Specifically, three images
(relax, move left arm, move right for one participant; relax, move left
arm, mental cube rotation) were translated into different robot commands
by the BCI, with the exact interpretation of the mental state depending
on the location of the robot. For example, if the robot was located in an open
area, the “move left arm” motor image caused the robot to turn left; how-
ever, if there was a wall to the robot’s left, “move left arm” caused the robot
to follow the wall. In all situations, the “relax” image caused the robot to
move forward and automatically stop when an obstacle was detected in front
of it. Finally, three lights on top of the robot were always visible to the par-
ticipants and indicated which of the three motor or mental images was cur-
rently being detected by the BCI. Using this control approach, the two
participants were able to complete steering and navigational tasks nearly