1. A Robust Layered Control System for a Mobile Robot Rodney A. Brooks Presenter: Michael Vidal

2. Requirements Multiple Goals Multiple Goals Multiple Sensors Multiple Sensors Robustness Robustness Extensibility Extensibility

3. Starting Assumptions 1. Complex behavior need not be reflected by a complex control system 2. Things should be simple 3. Map making is of critical importance 4. A robot must model in 3 dimensions 5. Relational maps are more useful 6. Build no artificial environment for the robot

4. Starting Assumptions 7. Visual data allows for intelligent interaction with the world 8. Processing steps should be self- calibrating 9. Robots should be self-sustaining

5. Levels of Competence Typical composition of robotic functions includes 5 sections: Typical composition of robotic functions includes 5 sections: 1. Sensing 2. Mapping 3. Planning 4. Task execution 5. Motor Control This division represents a solution based on the internal functionality of the robot This division represents a solution based on the internal functionality of the robot

6. Levels of Competence Proposed approach creates levels based on expected external functionality Proposed approach creates levels based on expected external functionality 1. Avoid contact with objects 2. Wander around aimlessly (without hitting things) 3. “Explore” the world 4. Build a map of the world to plan routes 5. Notice changes in the “static” environment 6. Reason about the world and perform tasks 7. Formulate and execute plans to change the world 8. Reason about the behavior of objects and modify plans accordingly

7. Layers of Control Layers of control directly map to layers of confidence Layers of control directly map to layers of confidence Subsumption architecture Subsumption architecture Build and debug a control layer with level 0 confidence Build and debug a control layer with level 0 confidence Level 0 layer should never be altered Level 0 layer should never be altered Add one layer of control at a time to previous layers Add one layer of control at a time to previous layers A layer may examine the data of the layer below it A layer may examine the data of the layer below it A layer may interfere with the layer below it A layer may interfere with the layer below it

8. Layers of Control This approach naturally lends itself to meeting the stated requirements This approach naturally lends itself to meeting the stated requirements Multiple Goals: Individual layers may work on individual goals concurrently Multiple Goals: Individual layers may work on individual goals concurrently Multiple Sensors: Sensors need not feed data into some central representation Multiple Sensors: Sensors need not feed data into some central representation Robustness: Lower layers continue to function when higher layers fail Robustness: Lower layers continue to function when higher layers fail Extensibility: Each layer can run on its own processor Extensibility: Each layer can run on its own processor

9. Layer Structure Each layer (module) an individual processor Each layer (module) an individual processor Each module has some number of inputs and outputs Each module has some number of inputs and outputs Modules connected by “wires” Modules connected by “wires” “Wires” generally connect a layer’s output to the input of the layer below “Wires” generally connect a layer’s output to the input of the layer below Messages passed are unreliable Messages passed are unreliable

10. Implementation 0-Level Layer: “Avoid” 0-Level Layer: “Avoid” Ensures that the robot does not come in contact with other objects Ensures that the robot does not come in contact with other objects Will avoid stationary objects Will avoid stationary objects Will flee from moving obstacles Will flee from moving obstacles Consists of a number of mini-modules, including “sonar”, “collide”, “feelforce”, “runaway”, “turn”, and “forward” Consists of a number of mini-modules, including “sonar”, “collide”, “feelforce”, “runaway”, “turn”, and “forward” The latter two interact directly with the robot The latter two interact directly with the robot

11. Implementation Level 1 Layer – “Wander” Level 1 Layer – “Wander” Creates a new destination for the robot every few seconds Creates a new destination for the robot every few seconds Relies on 0-level functionality to avoid obstacles Relies on 0-level functionality to avoid obstacles Adds two mini-modules to the system: “Wander”, and “Avoid” Adds two mini-modules to the system: “Wander”, and “Avoid”

12. Implementation Level 3 Layer: “Explore” Level 3 Layer: “Explore” Allows the robot to seek out interesting places to visit Allows the robot to seek out interesting places to visit Adds the mini-modules “Stereo”, “Look”, “Pathplan”, “Integrate”, and “Whenlook” Adds the mini-modules “Stereo”, “Look”, “Pathplan”, “Integrate”, and “Whenlook” Impedes output of level 1 layer to reach its goal Impedes output of level 1 layer to reach its goal