282
S. T. Nichols et al.
If the neural response increases, then the agent continues to turn in that
same direction. The agent’s notion of where the stimulus is located depends on
the ratio
major ef f
minor ef f
. The more aligned the agent’s orientation is with the stimulus
(i.e. heading straight towards the stimulus), the greater the ratio
major ef f
minor ef f
(i.e.,
closer to an ellipse). The larger
major ef f
minor ef f
is, the better the agent knows which
direction the stimulus is coming from. In contrast, stimuli frequencies that generate lower values of
major ef f
minor ef f
(i.e., closer to a circle) make it harder for the agent
to determine the direction it should go to reach the goal.
Additionally, we use the ratio
major ef f
minor ef f
to compute a variance that introduces
error into the directed motion of the agent. The larger the ratio, the lower
the variance, resulting in less motion error. The lower the ratio, the higher the
variance, resulting in greater motion error (Fig. 2 - panels 4 and 5). The model
uses a uniform distribution so there is an equal chance that the agent chooses
any direction in this range of turn angles. Once the behavioral algorithm chooses
a sign and direction, the agent turns accordingly and continues to move until it
reaches the goal.
2.5 Navigational Environment
To match the environment of [8], the agent is modeled as a circle with a radius
of 1.1 cm to match the average length of a leech, 2.2 cm. The target location
is a circle with a radius of 7 cm. The agent must stay in the target location
for at least 30 time-steps before it is counted as a successful find. The starting
location is (x = 13.00 cm, y = 13.00 cm), and the goal location is (x = 65.90
cm, y = 65.90 cm) (see Figs. 4Aiii, 4Biii, and 5Aiii for example trajectories).
Although the agent is currently able to move to any location in this study, we
plot the agent’s trajectories in a circle with a radius of 41.5 cm to match the
arena setup of [8]. Preliminary studies in which we constrained the agent to
only move within the arena (performed after submission of this work) 1) negate
some of the potential non-physicality of our model, and 2) do not show any
substantial performance degradation or qualitative result differences, and in fact
show slightly improved find-rates, leading us to believe that the approach used
in the present study is a valid starting point.
2.6 Experimental Setup
Simulations were performed to test the agent’s ability to find a target location
using the stimuli frequencies from [8]. The frequencies for mechanical and visual
inputs were 2 Hz, 4 Hz, 8 Hz, 12 Hz, 20 Hz, and 24 Hz. Each experiment was
performed using one hundred trials. For each frequency, the average find rate
was computed over all the trials. Experiments were done using only mechanical
stimulus, only visual stimulus, and combination of both stimuli (i.e., multimodal
stimuli). For the multimodal experiment, both stimuli were generated from the
same target location at the same exact frequencies.
S. T. Nichols et al.
If the neural response increases, then the agent continues to turn in that
same direction. The agent’s notion of where the stimulus is located depends on
the ratio
major ef f
minor ef f
. The more aligned the agent’s orientation is with the stimulus
(i.e. heading straight towards the stimulus), the greater the ratio
major ef f
minor ef f
(i.e.,
closer to an ellipse). The larger
major ef f
minor ef f
is, the better the agent knows which
direction the stimulus is coming from. In contrast, stimuli frequencies that generate lower values of
major ef f
minor ef f
(i.e., closer to a circle) make it harder for the agent
to determine the direction it should go to reach the goal.
Additionally, we use the ratio
major ef f
minor ef f
to compute a variance that introduces
error into the directed motion of the agent. The larger the ratio, the lower
the variance, resulting in less motion error. The lower the ratio, the higher the
variance, resulting in greater motion error (Fig. 2 - panels 4 and 5). The model
uses a uniform distribution so there is an equal chance that the agent chooses
any direction in this range of turn angles. Once the behavioral algorithm chooses
a sign and direction, the agent turns accordingly and continues to move until it
reaches the goal.
2.5 Navigational Environment
To match the environment of [8], the agent is modeled as a circle with a radius
of 1.1 cm to match the average length of a leech, 2.2 cm. The target location
is a circle with a radius of 7 cm. The agent must stay in the target location
for at least 30 time-steps before it is counted as a successful find. The starting
location is (x = 13.00 cm, y = 13.00 cm), and the goal location is (x = 65.90
cm, y = 65.90 cm) (see Figs. 4Aiii, 4Biii, and 5Aiii for example trajectories).
Although the agent is currently able to move to any location in this study, we
plot the agent’s trajectories in a circle with a radius of 41.5 cm to match the
arena setup of [8]. Preliminary studies in which we constrained the agent to
only move within the arena (performed after submission of this work) 1) negate
some of the potential non-physicality of our model, and 2) do not show any
substantial performance degradation or qualitative result differences, and in fact
show slightly improved find-rates, leading us to believe that the approach used
in the present study is a valid starting point.
2.6 Experimental Setup
Simulations were performed to test the agent’s ability to find a target location
using the stimuli frequencies from [8]. The frequencies for mechanical and visual
inputs were 2 Hz, 4 Hz, 8 Hz, 12 Hz, 20 Hz, and 24 Hz. Each experiment was
performed using one hundred trials. For each frequency, the average find rate
was computed over all the trials. Experiments were done using only mechanical
stimulus, only visual stimulus, and combination of both stimuli (i.e., multimodal
stimuli). For the multimodal experiment, both stimuli were generated from the
same target location at the same exact frequencies.