Sensors & Perception
Exteroceptive Sensors
This is a class of sensors that measure information about a robot’s external environment. They are characterised by a number of different attributes:
Field of View & Range
Every sensor has a region of space that it can cover:
- The width of that region is the field of view $\beta$
- The range $R$ is how far the field of view extends.
Sensor Performance
- Sensitivity
- Measure of the degree to which incremental changes in the target input change the output signal.
The ratio of output change to input change.
- Cross-sensitivity
- Sensitivity to environmental parameters that are orthogonal to the target parameters.
An example is that a compass is sensitive to magnetic north, but also to ferrous materials.
Accuracy & Precision
- Accuracy (Error)
- Difference between the sensor’s measured value, $m$ and the true value $v$, such that: \[\begin{align} \text{error}&=m-v\\ \text{accuracy}&=1-\frac{\lvert m-v\rvert}v \end{align}\]
- Precision
- Reproducibility sensor results. If the random error of a sensor is characterised by some mean $\mu$ and standard deviation $\sigma$, the the precision is the ration of the sensors output range to $\sigma$.
Systematic & Random Error
Systematic errors are deterministic:
- Caused by factors that can be modelled and corrected.
Random errors are non-deterministic:
- No prediction is possible.
- They can be described probabilistically, like errors in wheel odometry.
Resolution, Linearity & Frequency
- Resolution
- The minimum difference between two values.
- Bandwidth (Frequency)
- The speed which which a sensor can provide readings.
Responsiveness
Most sensors work poorly in certain domains. Therefore we should use sensors that have a good signal-to-noise ratio in their application.
For example, sonar works poorly in environments with large amounts of glass due to the large amount of specular reflection.
Behavioural Sensor Function
- Sensor Fusion
- The combination of information from multiple sensors into a single percept.
We want to use multiple sensors for the following reasons:
- Redundancy
- Complementary - Difference percepts that support each-other.
- Touch sensors when objects are too close to be detected by sonar.
- Coordinated - Using one sensor as a consequence of a percept detected by another.
- Detecting a possible obstacle using a range sensor and verifying it’s existence using a tactile sensor.
Observed Feature | No Feature Observed | |
---|---|---|
Feature Exists | True Positive | False Negative |
Nothing to Perceive | False Positive | True Negative |
Sensor Fusion
Sensor fusion ca be combined with behaviours in different ways:
- Sensor Fission
- One sensor per behaviour.
- Behaviours can share sensor streams.
graph LR s1[sensor] -->|percept| b1[behaviour] -->|action| cm[combination mechanism] s2[sensor] -->|percept| b2[behaviour] -->|action| cm[combination mechanism] s3[sensor] -->|percept| b3[behaviour] -->|action| cm[combination mechanism] cm -->|action| x(( ))
- Action-Oriented Sensor Fusion
- The combination of sensor data may trigger different different behaviours.
- An abstract percept emerges form the fusion of several percepts.
graph LR s1[sensor] -->|percept| fusion s2[sensor] -->|percept| fusion s3[sensor] -->|percept| fusion fusion -->|percept| behaviour behaviour -->|action| x(( ))