# Automation vs Autonomy ###### tags: `CS233` `robotics` `presentation` --- Q: What do Automation and Autonomy mean to you? --- ## Automation vs Autonomy - Automation: Robot as a tool - Preplanned motions - Deterministic algorithms - Understood environment - Autonomy: Robot as an agent - Plan generation - Symbol manipulation - Open world --- ### Autonomous Capability Sliders  ---- Discuss these robots on the slider scale: 1. Roomba 2. Our robot 3. Shakey 4. Unimate 5. Genghis 6. DaVinci 7. Deep Space One 8. Predator 9. Remus --- #### Plans - Automation: Execute pre-programmed actions or behaviors. Control theory. - Autonomy: Generate plan of action based on current world knowledge. Artificial Intelligence. --- #### Actions - Deterministic: Given same inputs, results in same output/action. - Non-deterministic: Many possible outputs given seemingly same inputs. - Some times caused by noisy or numerous inputs - Example: Walking down a corridor - Differences in testing: - Deterministic: Can prove correctness - Non-deterministic: statistical - [Genghis robot](https://en.wikipedia.org/wiki/Genghis_(robot)) --- #### World Models - World model: How the robot understands the world around it - Spatial (e.g. Maps) - Social (beliefs, desires of other agents) - May be preprogrammed or learned or both - Closed world: everything possible is known - Open world: list of possible states/objects is not completely specified - Frame problem: Need to make assumptions about the world. - What is not changing? - What is relevant? --- #### Knowledge Representation - Signals: Direct sensory inputs - Symbols: Higher level concepts (objects, walls) - Converting signals to labeled symbols is hard --- #### Bounded rationality - Autonomy in robots not the same as humans - Humans free to do as they choose - Robots are "mechanically autonomous" - Autonomous robots are still limited by the parameters of their programming --- #### Functional failures - "Tunnel vision": expectations based on incorrect closed world assumptions - "Substitution myth": Robots are not as adaptable as humans, and can't easily replace humans - Human out of the loop problem --- #### Design trade-offs for autonomous capabilities - Fitness to the task: Optimality vs resilience - Plans: Efficiency vs thoroughness - Impact: Centralized vs decentralized - Perspectives: Local vs global. Fly in the window effect - Responsibility: Who is responsible for reaching goals?