Updated: Mar 25
The performance and its definition that a robot focuses on are listed below.
Field of view (FOV) defines the angular area of the perceivable field of the sensor.
Density means the angular step size between sample points. It can be different horizontally and vertically.
Resolution equals the multiplication of FOV and Density generally.
Depth Resolution is the step size between possible measurement ranges along the measurement axis.
Depth Accuracy means the difference between measured range and actual range. It should be noted that the limitation of depth accuracy depends on the depth resolution. The highest depth accuracy will NOT exceed the depth resolution. While depth accuracy may depend on the subjective surroundings such as lighting and background, depth resolution is an objective metric of a sensor.
Minimum and Maximum Range defines the distances that are perceivable to the sensor. This may vary by object material, brightness, reflectivity, and the ambient environment.
Data Rate might be the frame rate or point rate, depending on the modality of the sensor.
Latency is the timing difference between capturing by the sensor and receiving by the backend SoC.
The primary function that a robot shall have to complete a task and the critical factor to this function.
High frame rate, depth accuracy, and low latency are required to provide data timely to avoid obstructions.
The frame rate, depth accuracy, and latency requirement depend on how fast a robot moves.
For example, a home-use cleaning robot travels at an average speed of 20 to 30 cm per second, but an AMR may travel at a speed up to 100cm per second.
A robot shall be able to distinguish foreground objects from the background. In certain scenarios, the requirements of the resolution depend on the minimum detectable size of the object.
A cleaning robot, for example, shall be able to distinguish a cable at a certain range; very often an AMR (autonomous mobile robot) must be capable of detecting or recognizing human for safety issue.By the way, a service robot, on the other hand, will integrate facial recognition and/or gesture control to enhance its functionality.
The function of volume measurement varies with different robots.
The accuracy of such volumetric data is generally at a fair level.
For example, a factory robot uses volumetric data to improve the efficiency in packaging, placement, or storage; a retail robot measures the space of the empty shelf for inventory restocking; a service robot uses volumetric data to properly deliver the carts.
SLAM is the abbreviation of “Simultaneous Localization and Mapping,” which allows a robot to quickly calibrate its location to fetch, travel, and self-navigate within a space.
SLAM requires fair to high depth accuracy and, similar to collision avoidance, the frame rate and latency requirement depends on the moving speed of the robot.
It is worthy to note that the map of a space constructed by stereo vision benefits the A.I. algorithm to do the scene understanding, allowing a robot to be adaptive to a versatile environment.
How Functions Change Criteria Weighting of Sensor Performance
Below is the matrix showing the weighting on the sensor performance related to the primary function stated above. However, please be noted that this is only a qualitative evaluation rather than quantitative evaluation. The actual weighting may vary case by case.