Skip to main content
Figure 10 | Robotics and Biomimetics

Figure 10

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

Figure 10

Illustration of the vector field for the coastal and GMM policy. Top left: Coastal policy. There is only one possible direction for every state at any time. The values of λ2 in the cost function were set experimentally. Others: The GMM policy for three different levels of uncertainty. For each point, multiple possible actions are possible which are reflected by the number of arrows (only the first three most likely actions). As the uncertainty decreases, the policy becomes less multi-model, but still is around the edges and corners. Note that once being certain if one is close to the edge there is a possibility to go either straight to the goal or stay close to the edge and corners.

Back to article page