To plan, or to respond? How should we build robots? Should they be programmed to look ahead and to strategize, or should they be reflexive and respond to their environment? This unit begins to look at the ramification of either strategy.
Table of Contents
Reading – Chapter 13: Think Hard, Act Later
Planning behaviours are also known as deliberative architectures or SPA for sense, plan, act. The methodology uses sequential functional modules to collect information, process it and then act. It is a centralized reasoning approach that can be slow due to the amount of data processing needed for each action.
Food For Thought 13-1
Can you use deliberative control without having some internal representation?
I would say no. The deliberative control system is also known as sense. plan, act. In order to plan some sort of representation of the problem is needed. The robot collects information about its enviroment, and then plans how it can best achieve its goals in that environment. This “understanding/evaluation” of the environment necessitates an internal representation.
Food For Thought 13-2
Can animals plan? Which ones, and what do they plan?
Yes, I do believe that animals can plan. They can learn about their environment and then refine problem solving based upon experience. I’d recommend watching this YouTube video by Mark Rober to see how squirrels learn how to solve a maze.
Food For Thought 13-3
If you had perfect memory, would you still need to plan?
Yes, because even though we can have a million plans stored away, the world changes. When there are unplanned changes we need to create a new plan to respond to the new situation. There’s a saying the best laid plans of mice and men, or no plan survives contact with the enemy. These sayings imply that no matter how good our plans, they will have to adapt to changing situations.
Reading – Chapter 14: Don’t Think, React!
Reactions are based upon predetermined responses to stimuli. The process bypasses the lengthy process of planning and couples sensing with action. These architectures decompose problems and responses into base responses that are built upon each other so that the most basic tasks are processed first while additional response is built upon existing responses.
Food For Thought 14-1
Can you change the goal of a reactive system? If so, how? If not, why not? You will soon learn how other methods for control deal with this problem and what the trade-offs are.
I would say that reactive systems cannot have their goals dynamically adjusted. The system links predetermined responses to expected inputs. Additional layers can be built upon lower layers to create increasing complexity, but the base layers will continue to be processed with the highest priority. The only way that the goal of a reactive system could be adjusted would be to rewrite the base layers of the logic, which would be more akin to a new program than a shifted goal.
Food For Thought 14-2
Can you always avoid using any representation/world model? If so, how? If not, why not, and what could you do instead?
It depends on the complexity of the task that you would like to accomplish. Some simple or predictable tasks will be able to operate without planning or representation. While other more complex tasks will require some planning and problem solving to accomplish. For example the early examples of tortoises that we discussed were able to achieve a goal and display behavior without planning or internal representation. They depended only on basic sense and response to be able to operate and eventually locate their charging base.
Food For Thought 14-3
Can a reactive robot learn a maze?
No. Because there is no representation, the robot cannot remember any solutions that it might discover. Everytime the robot enters the maze it will need to use the same sense, act loop to navigate the maze and will be unable to optimize its route based upon previous attempts.
Conclusion
In this unit we began to look at some simplified approaches to robotic control and behavior. The two methods we discussed are virtually as opposed as it is possible to be, and while both have advantages both have disadvantages as well. In the next unit I believe we will begin to look into hybrid approaches that are even better than these.
There are no exercises listed in the instructors notebook for this or any of the following sections. The next code and wiring examples are likely to be when I am working on my final project after I finish these units.
Shawn Ritter
November 29th, 2021
Featured Image: Photo by Andrea Piacquadio: https://www.pexels.com/photo/close-up-photography-of-squirrel-932343/