Skip to main content
Figure 9 | Robotics and Biomimetics

Figure 9

From: Learning search polices from humans in a partially observable context

Figure 9

Illustration of three different types of modes present during the execution of the task. The robot is being controlled by the learned GMM model. The white ball represents the actual position of the robot’s end effector. The blue ball represents the believed position of the robot’s end effector and the robot is acting according to it. The blue ball arrows represent modes. Colours encode the mode’s weights given by the priors π k after conditioning (but not re-weighted as previously described). The spectrum ranges from red (high weight) to blue (low weight). Top left: Three modes are present, but two agree with each other. Top right: Three modes are again present indicating appropriate ways to reduce the uncertainty. Lower left: Two modes are in opposing directions. No flipping behaviour between modes occurs since preference is given to the modes pointing in the same direction as the robot’s current trajectory. Lower right: GMM modes when conditioned on the state represented in the lower left figure. The two modes represent the possible directions (un-normalised).

Back to article page