The disorientation of the human subject is to ensure that his believed location is uniform.
In Experiment 1, after passing the FB task, participants were asked what a puppet would do after he failed to find his toy at the falsely believed location.
The second initialisation, #2, reflects the situation where once again their is a large distance between true and believed location of the eof.
The red cluster always has the highest likelihood which is taken to be the believed location of the robot's/human's end effector.
The circle with a black dot at its centre indicates the true starting point of the end effector (eof), whilst the triangle with the black dot is the initial believed location of the eof.
When the anchors' coordinates are known up to a confidence level, they can be modeled as ordinary sensors, connected to an imaginary anchor (with infinite mass and set at the believed location) by a hard spring, with strength k much higher than others, reflecting the confidence level of the anchor in its coordinates.
The choice of the believed location of the robot/human end effector is taken to be the mean of the Gaussian function with the highest weighted π. x ̂ t = arg max x t p u x t | z 0 : t ; π, μ, Σ = μ ( k = max (6). Figure 3 depicts different configurations of the modes (clusters) and believed position of the end effector (indicated by a yellow arrow).
In the first experiment, the true and believed locations of the end effector were drawn uniformly from the original start distribution (Figure 2, top right) reflecting the default setting.
In experiment #2, both greedy and coastal policies perform equally well and will usually perform faster than the GMM model if the true and believed locations of the end effector do not leave the surface of the table.
My cofounder, Matt Galligan, and I firmly believe that location is in a similar position as social was in 2001 or so.
