Bio-inspiration in Embodied AI: Perception and action Flashcards
What is biologically-inspired robotics? Can you give an example?
Animals as inspiration for robotic design:
Example: Study how the brain of an ant control its 6 legs in order to walk the way it does => Build an ant-like robot.
- Do not need to copy everything, but robot has to face the same challenge as the biological system (the principle you are studying)
Why is inspiration from biology useful?
Animals are experts in specific motion (cheeta good at running), where humans are not.
What is sensing/perception?
Sensing: What is detected by the sensory system (e.g., what we see = light)
Perception: Process of selecting, organizing and interpreting the sensations (how we understand/interpret the sensory information: red apple)
What is the difference between biological perception and artificial perception? Can you give an example of both?
Biological perception: Looking one item at a time
- Example: See one apple at a time on an apple tree. Looking at the closest apples first before looking at partially occluded apples.
Artificial perception: Looking at one pixel at a time
- Example: Use computer vision to process the picture of an apple tree, to detect where the apples
How does movement help perception? Can you give an example?
Get sensor data from different viewpoints => Better perception
Example: Picking apples from an apple tree.
Walking around the apple tree => better see occluded apples.
What is active sensing? Can you give an example of active sensing in nature/machines?
The act of moving the sensors to achieve better perception of the environment and updating/correcting the plan while it is being executed.
Example Nature: Snow Fox
What is motion perception? Can you give an example?
It is the process of estimating the speed and direction of elements/objects in a scene based on sensory information (the direction and speed of a car is driving in the scene)
How does motion detection work in the brain?
Human ‘detector circuit’ uses relative movement:
○ Neurons only generate a signal if something has changed => detect motion - i.e. something in the scene has changed.
○ When the scene changes => each neuron generates a signal when a change is detected (light changes as the light moves).
○ The brain has a mechanical way of delaying signals from multiple neurons to a neuron (i.e. an AND gate: only generate a signal if something has changed => detect motion - i.e. something in the scene has changed).
How are perception and action related to each other?
- Intelligent perception can help generate intelligent actions
- Intelligent action can lead to intelligent perception
