Bioinspired Navigation Based on Distributed Sensing in the Leech
285
3 Results
Figure 4A shows the results of our model for the mechanical experiments, along
with a comparison to data observed in real animals from [8], and an example
trajectory. With the exception of 2 Hz, both the model and animal experiment
data exhibit fairly good agreement, with a peak find rate of 0.5 at 12 Hz.
Figure 4B shows the results of our model for the visual experiments, along
with a comparison to the data observed in real animals from [8], and an example
trajectory. The find rates from model and the animal experiment data follow
the same overall trends, with peak find rates occurring at 2 Hz, and find rates
decreasing with increasing frequency. We note here that our model currently
over-predicts the find rate (see Sect. 4 for details).
Figure 5A shows the results of our model for the multimodal experiments,
along with a comparison to the data observed in [8], and an example trajectory. The find rates in the multimodal experiment are higher for all frequencies than for either of the visual or mechanical stimui alone, demonstrating the
elliptical model’s ability to effectively combine multiple sensory inputs. However, the trends observed in our multimodal experiments reflect the trends in
our visual experiments, suggesting that our model currently overly-weights the
visual response (see Sect. 4).
4 Discussion and Conclusions
Overall, our model demonstrates fairly good agreement with the animal observations found in [8], though there are areas where it can be improved. In the
mechanical experiments, an anomaly in the find rate trend occurred at 2 Hz
where the modeled find rates came out to be substantially higher than the actual
results. [10] showed a similar outlier in spike rate measurements, and stated that
it did not correlate with the behavioral results found. Other differences in data
occur at the higher end of the stimuli frequencies. The modeled find rates at
20 Hz and 24 Hz appeared lower than the actual data, suggesting that higher
frequencies may elicit a different kind of motor behavior even if it does not create larger spike rates (Fig. 4Ai and Aii). For our visual experiments, while the
trends between the modeled and observed data were the same, the peak find
rates in the lower frequency range of the modeled data are higher than those
of the animal experiment data. Better fits might be obtained by altering the
model parameters of the visual part of the elliptical model, and modifying the
variance rate. The higher frequencies elicited accurate find rates, although some
of the trends do not line up with the animal experiment find rates. With additional tuning of our model, we believe that these discrepancies can be removed.
Alternatively, this could be due to the fact that we only modeled the frequency,
but not intensity of the visual stimulus for simplicity. In reality, the visual stimulus also depends on intensity. Including this effect would likely improve our
results. Data from the multimodal experiments resemble the trends seen in the
visual results, suggesting that our model currently over-weights the effects of
285
3 Results
Figure 4A shows the results of our model for the mechanical experiments, along
with a comparison to data observed in real animals from [8], and an example
trajectory. With the exception of 2 Hz, both the model and animal experiment
data exhibit fairly good agreement, with a peak find rate of 0.5 at 12 Hz.
Figure 4B shows the results of our model for the visual experiments, along
with a comparison to the data observed in real animals from [8], and an example
trajectory. The find rates from model and the animal experiment data follow
the same overall trends, with peak find rates occurring at 2 Hz, and find rates
decreasing with increasing frequency. We note here that our model currently
over-predicts the find rate (see Sect. 4 for details).
Figure 5A shows the results of our model for the multimodal experiments,
along with a comparison to the data observed in [8], and an example trajectory. The find rates in the multimodal experiment are higher for all frequencies than for either of the visual or mechanical stimui alone, demonstrating the
elliptical model’s ability to effectively combine multiple sensory inputs. However, the trends observed in our multimodal experiments reflect the trends in
our visual experiments, suggesting that our model currently overly-weights the
visual response (see Sect. 4).
4 Discussion and Conclusions
Overall, our model demonstrates fairly good agreement with the animal observations found in [8], though there are areas where it can be improved. In the
mechanical experiments, an anomaly in the find rate trend occurred at 2 Hz
where the modeled find rates came out to be substantially higher than the actual
results. [10] showed a similar outlier in spike rate measurements, and stated that
it did not correlate with the behavioral results found. Other differences in data
occur at the higher end of the stimuli frequencies. The modeled find rates at
20 Hz and 24 Hz appeared lower than the actual data, suggesting that higher
frequencies may elicit a different kind of motor behavior even if it does not create larger spike rates (Fig. 4Ai and Aii). For our visual experiments, while the
trends between the modeled and observed data were the same, the peak find
rates in the lower frequency range of the modeled data are higher than those
of the animal experiment data. Better fits might be obtained by altering the
model parameters of the visual part of the elliptical model, and modifying the
variance rate. The higher frequencies elicited accurate find rates, although some
of the trends do not line up with the animal experiment find rates. With additional tuning of our model, we believe that these discrepancies can be removed.
Alternatively, this could be due to the fact that we only modeled the frequency,
but not intensity of the visual stimulus for simplicity. In reality, the visual stimulus also depends on intensity. Including this effect would likely improve our
results. Data from the multimodal experiments resemble the trends seen in the
visual results, suggesting that our model currently over-weights the effects of