# Software Architectures ###### tags: `CS233` `robotics` `presentation` --- Warmup activity: You need to program a robot to navigate a maze. - What are all the different "functions" you may need? - What assumptions are you making about the nature of the maze?  --- ## Types of software architectures - Operational: describe what the system does - Systems: Describe how the system is decomposed - Technical: Describe implementation details --- ## Operational architectures for robots - Described in terms of *layers* - Which function goes to which layer depends on five attributes: - *primitives* - *perceptual ability* - *planning horizon* - *time scale* - *use of models* --- ### Primitives - Four primitives: SENSE, PLAN, ACT and LEARN - SENSE: takes sensor input and produces output useful to other functions - PLAN: takes information from sensors or world model, and produces tasks - ACT: produces commands to the motors/actuators - LEARN: mechanisms for improving the other operations --- | Robot primitives | Input | Output | |:---------------- |:-------------------------- |:----------------- | | SENSE | Sensor data | Sensed info | | PLAN | Info (sensed or cognitive) | Directives | | ACT | Directives or sensed info | Actuator commands | --- Practice. Classify the following operations: 1. Turn head 30 degrees to the right 2. Detect obstacle ahead 3. Choose between fleeing mode and roaming mode 4. Determine distance to object 5. Decide that when you encounter a wall you will move to the right along it 6. Play a sound 7. Play a sound that depends on the distance to an object --- ### Perceptual ability - Perception is the conversion of sensory input into information - Two kinds of perception: - Direct perception, directly based on the incoming signals - Recognition, based on the conversion of sensory information into symbols --- ### Planning horizon What data does the capability need - present - past and present - future, past and present --- ### Time scale How quick must the functionality be: - very fast - fast - slow --- ### World model - local world model: sensor information collected just for that function - global world model: sensor information is first processed into a global view --- ## AI Operational architecture layers  --- ## Systems architecture layers  --- ### Systems architecture paradigms Can be clasified based on two traits: - Interaction between primitives - Route of sensing --- Interactions: - SENSE, PLAN, ACT (in sequence) - SENSE-ACT (coupled) - PLAN, SENSE-ACT --- Sensing routes: - local sensing: information goes directly to function that needs it - global sensing: information is first compiled into the world model - hybrid sensing: information goes in both directions --- ### Systems architecture paradigms | Paradigm | Primitives | Sensing | |:------------ | ---------------- | ------- | | Hierarchical | SENSE, PLAN, ACT | Global | | Reactive | SENSE-ACT | Local | | Hybrid | PLAN, SENSE-ACT | Hybrid |