Skip to main content
Figure 7 | Robotics and Biomimetics

Figure 7

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

Figure 7

Risk of searches. Illustration of risk-prone and risk-averse searches in terms of a risk factor (left) and expected sensation (right). Left: Each trajectory was reduced to a single scalar, which we call the risk factor, quantizing the risk of a trajectory. The risk factor is inversely proportional to the sum of the information gain of a particular trajectory. The colour paired dots (risk averse) and squares (risk prone) represent trajectories which are plotted in Figure 8, to illustrate that these correspond to risk-averse and risk-prone searches. Right: Corresponding trajectories chosen in the risk factor space but represented in the feature space. As expected, trajectories with a high risk map to regions of low expected feature. However, the transition from the risk space to feature space is non-linear and will result in a different risk-level classification than the feature metric previously discussed.

Back to article page