Limits...
Information-Driven Active Audio-Visual Source Localization.

Schult N, Reineking T, Kluss T, Zetzsche C - PLoS ONE (2015)

Bottom Line: These actions by the robot successively reduce uncertainty about the source's position.Because of the robot's mobility, this approach is suitable for use in complex and cluttered environments.We present qualitative and quantitative results of the system's performance and discuss possible areas of application.

View Article: PubMed Central - PubMed

Affiliation: Cognitive Neuroinformatics, Bremen University, Bremen, Germany.

ABSTRACT
We present a system for sensorimotor audio-visual source localization on a mobile robot. We utilize a particle filter for the combination of audio-visual information and for the temporal integration of consecutive measurements. Although the system only measures the current direction of the source, the position of the source can be estimated because the robot is able to move and can therefore obtain measurements from different directions. These actions by the robot successively reduce uncertainty about the source's position. An information gain mechanism is used for selecting the most informative actions in order to minimize the number of actions required to achieve accurate and precise position estimates in azimuth and distance. We show that this mechanism is an efficient solution to the action selection problem for source localization, and that it is able to produce precise position estimates despite simplified unisensory preprocessing. Because of the robot's mobility, this approach is suitable for use in complex and cluttered environments. We present qualitative and quantitative results of the system's performance and discuss possible areas of application.

No MeSH data available.


Entropy of the PDF estimate (robot experiments).The information-driven action selection process leads to a (roughly) exponential decrease of entropy. In comparison, random movements still lead to a mostly monotonic decrease of entropy, but at a slower rate and with a higher minimum entropy than in the information-driven approach.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4556528&req=5

pone.0137057.g011: Entropy of the PDF estimate (robot experiments).The information-driven action selection process leads to a (roughly) exponential decrease of entropy. In comparison, random movements still lead to a mostly monotonic decrease of entropy, but at a slower rate and with a higher minimum entropy than in the information-driven approach.

Mentions: Figs 10 and 11 show the results of the robot experiments. In Fig 10A, the average RMS error is plotted as a function of the number of actions, averaged over all experiments. The initial error after the first measurement is similar to the simulation results, showing that sensory measurements are of similar quality in simulation and in the robot scenario. After the first movement and subsequent visual and auditory measurements, the average RMS error is still 1.25m, which can be explained by the fact that, after one movement and two measurements, the system has limited distance information. After the second movement, the average error already decreases to less than 0.7m, and it continues to decrease to ca. 0.3m after the third movement. After the fourth movement, the average RMS error is below 0.2m, which already is smaller than the diameter of the source. Overall, the plot shows a smooth monotonic decrease of the average error, marginally worse than the results produced in the simulation environment. This is especially an interesting result because auditory measurement precision is worse than during simulation due to highly reverberant signals and the relatively simple template-matching-based vision system described above, which is principally prone to noise and illumination changes. Moreover, the combination of possible measurement errors for the initial position and the radius of the source also might affect performance negatively.


Information-Driven Active Audio-Visual Source Localization.

Schult N, Reineking T, Kluss T, Zetzsche C - PLoS ONE (2015)

Entropy of the PDF estimate (robot experiments).The information-driven action selection process leads to a (roughly) exponential decrease of entropy. In comparison, random movements still lead to a mostly monotonic decrease of entropy, but at a slower rate and with a higher minimum entropy than in the information-driven approach.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4556528&req=5

pone.0137057.g011: Entropy of the PDF estimate (robot experiments).The information-driven action selection process leads to a (roughly) exponential decrease of entropy. In comparison, random movements still lead to a mostly monotonic decrease of entropy, but at a slower rate and with a higher minimum entropy than in the information-driven approach.
Mentions: Figs 10 and 11 show the results of the robot experiments. In Fig 10A, the average RMS error is plotted as a function of the number of actions, averaged over all experiments. The initial error after the first measurement is similar to the simulation results, showing that sensory measurements are of similar quality in simulation and in the robot scenario. After the first movement and subsequent visual and auditory measurements, the average RMS error is still 1.25m, which can be explained by the fact that, after one movement and two measurements, the system has limited distance information. After the second movement, the average error already decreases to less than 0.7m, and it continues to decrease to ca. 0.3m after the third movement. After the fourth movement, the average RMS error is below 0.2m, which already is smaller than the diameter of the source. Overall, the plot shows a smooth monotonic decrease of the average error, marginally worse than the results produced in the simulation environment. This is especially an interesting result because auditory measurement precision is worse than during simulation due to highly reverberant signals and the relatively simple template-matching-based vision system described above, which is principally prone to noise and illumination changes. Moreover, the combination of possible measurement errors for the initial position and the radius of the source also might affect performance negatively.

Bottom Line: These actions by the robot successively reduce uncertainty about the source's position.Because of the robot's mobility, this approach is suitable for use in complex and cluttered environments.We present qualitative and quantitative results of the system's performance and discuss possible areas of application.

View Article: PubMed Central - PubMed

Affiliation: Cognitive Neuroinformatics, Bremen University, Bremen, Germany.

ABSTRACT
We present a system for sensorimotor audio-visual source localization on a mobile robot. We utilize a particle filter for the combination of audio-visual information and for the temporal integration of consecutive measurements. Although the system only measures the current direction of the source, the position of the source can be estimated because the robot is able to move and can therefore obtain measurements from different directions. These actions by the robot successively reduce uncertainty about the source's position. An information gain mechanism is used for selecting the most informative actions in order to minimize the number of actions required to achieve accurate and precise position estimates in azimuth and distance. We show that this mechanism is an efficient solution to the action selection problem for source localization, and that it is able to produce precise position estimates despite simplified unisensory preprocessing. Because of the robot's mobility, this approach is suitable for use in complex and cluttered environments. We present qualitative and quantitative results of the system's performance and discuss possible areas of application.

No MeSH data available.